“The Great Arsenal of Democracy”: The Political Economy of the County-Level Allocation of World War Two Military Spending Paul Rhode (UNC-CH) James Snyder (MIT) Koleman Strumpf (UNC-CH) July 2001 Highly Preliminary During the Second World War, the federal government assumed an unprecedented degree of control over the US economy. At the peak, the share of federal government expenditures in GNP soared to 44 percent, a level never attained before or since. In addition to enrolling 16.4 million Americans (that is, about one-eighth of the 1940 population) in the armed forces, the federal government spent $196 billion between June 1940 and June 1945 on military supply contracts and $31 billion on investments in new factories and military installations. In today’s purchasing power, this total would come to roughly two trillion dollars. Although this war effort represented the largest single intervention by the federal government in the economy, the political economy of these enormous spending flows has been subject to relatively little systematic scholarly investigation.1 This paper uses county-level economic and political data to investigate the determinants of the geographic allocation of World War II-era military spending, both for major war supply contracts and for new facility projects. It builds on two extensive literatures. The first, with important contributions by Reading (1973), Wright (1974), Wallis (1987, 1998. 2001), Anderson and Tollison (1991), Fleck (1999, 2001), Fishback et al. (2001) and others, investigates the political economy of the New Deal spending. The second is more contemporary, analyzing the geographic allocation of federal resources in the recent period (see Levitt and Snyder, 1995 and 1997 for an overview).2 Work that examines the contemporary sub-state distributions of funds generally focus on Congressional control or seniority. We extend this literature by evaluating the relative importance of a wider variety of political and non-political motives. As a caveat, World War Two was a period of great national crisis and our findings regarding the allocation of 1 See Rhode (2000) and Bateman and Taylor (2001a and 2001b) for analysis of state-level World War Two spending data. 2 Within this vast literature, Goss(1972), Mayer (1991), Ray (1981), Rundquist (1978), and Rundquist and Griffith (1976) focus specifically on the role of congressional power on the distribution of military contracts. As a quick summary, few of the studies find congressional committee assignments or seniority matter much. 2 funds may not readily generalize to periods of “normalcy.” Nonetheless, the wartime experience is of considerable interest in its own right. Our analysis focuses on two important sets of questions: (1) What was the relative importance of strategic, economic, and political forces in the spending decision process?3 (2) What were key determinants of spending within the political allocation process? For example, what was the relative importance of presidential versus congressional influence? Utilizing detailed data on the 3000 plus counties (as opposed to the 48 states) offers us much greater opportunity to disentangle these relationships. The larger sample size liberates us from the degrees of freedom problems that constrained some of the studies of the New Deal era (e.g. Anderson and Tollison (1991)) from including a full set of political and economic variables. Focusing on sub-state allocations provides tests that are not possible with state- level data. For example, many standard political allocation models suggest that funds should be disproportionately go to very large or very small states because of the Electoral College system while within each state, funds should be spread more evenly across large and small communities. Another benefit of using county-level spending is the ability to test for the importance of House committee assignments, leadership posts, and other member-specific and district-specific characteristics. Except in the big cities, county lines generally coincide with congressional district lines in the 1940s. The county units were geographically stable, making it is straightforward to calculate the degree of inter- temporal stability or instability of each county’s voting patterns (as in Wallis). County lines are almost surely exogenous (unlike, say, congressional districts), so there is no danger of grouping on an endogenous variable.4 Looking at the finer aggregate level also allows us to consider issues such as spillover effects of spending. 3 Using the state-level data, Rhode (2000) and Bateman and Taylor (2001b) both found that economic factors (chiefly, pre-war industrial capacity) rather than political factors (presidential voting behavior and congressional power) mattered most for the distribution of military purchases during World War Two. But Rhode also found political factors, especially party loyalty, did help explain where new investments in military and industrial facilities were located. These relationships may be clouded by state-level aggregation and using county-level data should produce more robust findings. 4 Disadvantages of using the county-level data include that fact that some counties are split between more than one congressional district; some of the manufacturing and election data are poorer quality at the county than state levels, and that there likely are meaningful economic and political spillovers across county lines. 3 This paper has the following form: the first section documents the unprecedented level of the federal government spending during World War Two and maps its highly uneven geographic distribution. The next section briefly traces the evolution of the government procurement policies and agencies over the war, highlighting accounts of the role of lobbying in shaping the allocation decisions. The third section extends the state- level political-economy model that is standard in the New Deal literature to encompass (a) the allocation of spending across jurisdictions within an election area; and (b) the joint role of Presidential and congressional candidates in determining spending outcomes. The fourth section discusses the implementation of this model and develops a baseline framework to incorporate strategic and economic determinants of spending. The fifth section presents our main empirical results and provides a series of “experiments” to interpret the magnitude of the estimated effects. The final section concludes, offering suggestions for future avenues of research. I Building Roosevelt’s “Great Arsenal for Democracy” required great effort from the American people. During the war, US industry produced 296 thousand aircraft, 76 thousand ships, 86 thousand tanks, 477 thousand bazookas, 20 million small arms (including rifles, machine guns, and hand guns), 27 million gas masks, 23 million ampules of sodium penicillin, 52 million pairs of boots and shoes, 271 million drawers, and much, much more.5 In current dollars, spending on military supply contracts, facilities, and project orders totaled $237.1 billion over the June 1940 to June 1945 period.6 In today’s (2000) dollars, this is equivalent to roughly $2,023 billion. Relative to the 1940 total population, per capita spending over this five-year period averaged $1,813 in current dollars or almost $15,500 in 2000 purchasing power. In real annual per capita terms, domestic procurement spending during World War Two was four-and-one- half times higher than the New Deal era spending which has attracted so much scholarly 5 War Production Board (1945) pp. 106-10. 6 Project orders are “orders for work issued to Government-owned arsenals, shipyards, manufacturing depots, and the like.” US Senate, Committee on National Defense Migration, Washington Hearings, p. 6582. 4 attention.7 Focusing strictly in fiscal terms (and ignoring its ability to make life-and- death decisions affecting the millions of service personnel), the federal government exercised unprecedented control over the US economy during the war. At the peak in 1944, the share of federal government expenditures in GNP reached 44 percent, many times the 4 to 7 percent share prevailing over the 1933-39 period.8 Defense spending accounted for over four-fifths of the federal government outlays. Table 1 presents data on the cumulative distribution of military procurement spending by category over time. Contract spending accounted for slightly over $196 billion, 86 percent of the total; facility projects (such as new or expanded military installations or munitions factories) received almost $32 billion. Of the contract spending, about 30 percent was for aircraft, 25 percent for ordnance, 15 percent for shipbuilding, and 5 percent for communication equipment. Of the facility investments, 57 percent was devoted to industrial projects, the remainder for military projects. These public funds accounted for roughly two-thirds of all investment in manufacturing plant and equipment during the war period. Some 1,595 new industrial plants were built using federal money as part of the war effort. Another important point drawn from Table 1 is that about one-half of total spending and about three-quarters of facility investments were allocated by the end of 1942, suggesting that many of the key decisions were made relatively early in the conflict. The WWII spending displayed a high degree of geographic variability. Table 2 provides information on the distribution of military spending (for contracts and facilities only) in the top twenty counties measured in total and per capita terms.9 The list for total 7 The total New Deal per capita spending between 1933 and 1939 was $363 according to the aggregated data for “expenditures, loans, and insurance” in Reading (1973) Table 1 divided by the 1930 total population. To adjust for inflation, we use changes in the GNP deflator between 1936 and 1943. 8 See Table 1 of the GDP accounts at http://www.bea.doc.gov/bea/dn/st-tabs.htm. 9 These figures apparently exclude expenditures for the Manhattan Project, which over the 1942-45 period cost $1,890 million dollars. (Hewlett and Anderson (1962) pp. 723-24.) Although the Los Alamos laboratory is the leading symbol of the atomic bomb program in the public mind, it absorbed relatively little of the money. Of the total expenditures, $1,188 million was allocated to projects/facilities at Oak Ridge (Anderson Co.) TN, $390 million to Hanford (Benton Co.) WA, $74 million to Los Alamos (Sandoval Co.) NM, and $27 million to heavy water plants (located in Trail, BC; Morgantown, WV; Montgomery, AL; and Dana, IN.) The total for the “two billion dollar gamble” was less than 1 percent of the expenditures analyzed in the paper. It appears unlikely that including the Manhattan Project spending would fundamentally change our results. From the available literature, the location criteria for the Manhattan project included finding remote, secure sites with good access to water and transportation. Oak Ridge and Hanford required access 5 spending includes the “usual suspects,” chiefly large industrial counties. Wayne County, MI (home of the Detroit automakers) is number one with 5.7 percent of total spending; Los Angeles, CA number two (with 4.9 percent); and Cook County, IL is number three (with 4.4 percent). Collectively the top twenty counties account for 39.3 percent of total spending. The top twenty list for per capita spending contains more surprises. Sarpy County, NB was number one with spending of $60,800 per 1940 resident, over 30 times the national average. This county, located just south of Omaha, had been home to the Army’s Offutt Air Field since the early 1920s and in 1940 became the site of Glenn Martin’s bomber assembly plant.10 In reviewing these data, it is worth emphasizing that almost 42 percent of the counties in our dataset (1285 out of the total 3072) received no major supply contracts or facility investments. To appreciate the enormous geographic variation of spending, it is helpful to examine Figures 1 through 4.11 (The four figures map, respectively, total spending, contract spending, facilities investment, and per capita spending at a county level.) The maps reveal that spending levels varied greatly across the nation, and even within virtually every state. Taking Nevada as an example, the figures show the southwestern counties received high levels of per capita spending whereas the northeastern counties to cheap electric power, and the New Deal dam-building programs in the Tennessee valley and the Pacific Northwest undoubtedly had an impact on the locational choices. A further requirement for the Hanford and Los Alamos sites were that no major population centers could be directly down-wind. As an aside, Los Alamos was chosen from among the New Mexico sites in part because Robert Oppenheimer wished to give his atomic scientists “a laboratory with a view.” Rhodes (1986) p. 450. The original Jemez Springs site was situated in a deep canyon which was judged “too confining.” The secrecy surrounding the project and its perceived importance in the war effort shielded the project from much congressional oversight. The House and Senate leadership and relevant committee chairs on both sides of the isle was kept informed about the “general state of the project” and of how the expenditures were hidden in the Appropriations bill. Despite its two-billion price tag and the need to build three government towns from scratch, most “members of Congress remained completely in the dark.” The Oak Ridge facility received its first oversight tour by “very small number of carefully selected legislators” in May 1945; the first congressional delegation to Hanford took place after V-J day. It is commonly asserted that President Truman first learned in detail about the atomic bomb in a 24 April 1945 letter from Secretary of War Henry Stimson. But it is clear that he was aware of the existence of the Manhattan project from his Senate defense investigating committee. Groves (1962) pp. 359-66. 10 The plant built B-26 Marauders and B-29 Superfortresses including both the “Enola Gay” and “Bock's Car” aircraft which the atomic bombs on Japan. See http://www.offutt.af.mil/geninfo/history/. 11 In some sense, procurement activity was widely distributed. There was at least one procurement office in every state and as far as our data permit us to determine every congressional district received some spending. To be more precise, every congressional district contained at least one county that received funding. Our county-level data do not permit use to investigate the allocation of money in counties split between districts. The wide distribution of activity undoubtedly reflects both politics and the vast scale and scope of the war effort. 6 received nothing. This diversity highlights the value of analyzing spending at the sub- state level. Nationally, several patterns stand out. Coastal areas generally received higher level of spending. Activity is also concentrated in the core industrial areas (including the Northeastern manufacturing belt, the Piedmont region in the south, and the Pacific Coast cities.) Lastly, there is a wide belt of counties in the Great Plains and northern Rockies without significant spending. It is notable that many of these states were among the high per capita spending areas during the New Deal. Six of the top ten states (North Dakota, South Dakota, Montana, Idaho, New Mexico, and Wyoming) during the New Deal period were in the bottom ten during the World War Two period.12 One of the notable differences between the war effort and most federal programs before or since is that the World War Two money was all-new and untied to state matching formulas. This simplifies the analysis because we do not have to focus on the strengths of pre-existing interest groups or on the decisions by state governments. Understanding how the vast new source of funds was allocated requires we turn our attention to the nation’s capital and investigate the evolution of procurement policy in greater detail. II In the two decades before 1939, the US government spent only one to two percent of national income on its military. Most of what little was spent for supplies and arms was allocated according to rigidly specified competitive procedures. Procurement officers in the Army or Navy would advertise for clearly defined quantities and qualities for a specific item, invited bids, and awarded the contract to the lowest qualified bidder. Long-term contracts and development/educational funds were rare. Even when a firm did receive R&D funding to create a design, the contract might be let to rival producer. In addition to using short-term, arms-length, competitive contracting, the federal 12 Including expenditures for the Manhattan project would raise New Mexico’s ranking, but apparently the state would still be in the bottom ten in per capita terms. Indeed, at the state-level, per capita spending over the two periods is negatively correlated—the correlation coefficient is –0.12 in the entire sample and –0.31 if Nevada is excluded. This observation suggests applying a unified political economy explanation to spending patterns in both periods will prove challenging. 7 government also imposed profit limits on aircraft and shipbuilding contracts under the 1934 Vinson-Trammel and 1936 Merchant Marine Acts.13 The desirably of military contracting was further reduced by bad experiences with contract cancellations following World War One and by public hostility towards the “merchants of death" as reflected in the 1934-36 investigations of the munitions industry by the Senate’s Nye committee. The chaotic World War One experience was not happy for the military either. In 1922, the Joint Army and Navy Munitions Board (ANMB) was formed to improve coordination of military procurement during periods of national emergency. In 1931 and roughly every two-three years thereafter, the ANMB issued an Industrial Mobilization Plan. The planners surveyed 25-30 thousand manufacturing plants, taking special notice of their machine tool capabilities and transportation situation, and devised their wartime munitions orders accordingly. The planners also assigned each plant to the Service branch it was to supply. When the third revision of the plan was completed in the spring/summer of 1939, over 10,000 plants were “given sealed orders to be opened in the event of war.” Kester, (1940) p. 684. The actual outbreak of full-scale war in September 1939 led to dramatic changes in “business as usual.” Table 3 offers a condensed timeline of the evolution of government agencies in charge of procurement and industrial mobilization over the 1939- 45 period. One key set of changes was the expediting acts of June 28 and July 2, 1940, which allowed negotiated, cost-plus-a-fixed-fee contracts and payment before delivery. While procurement authorities continued to use competitive bidding for small contracts, the vast majority of procurement contracts shifted to a negotiated basis.14 In October 1940, the federal government also eliminated profit ceilings on defense contracts, using excess-profit taxes in their place. Among the other important changes was establishment of a series of civilian-run bureaucracies to facilitate the war production effort. In May 1940, Roosevelt used powers dating back to the First World War to establish Advisory Commission of the Council for National Defense, which became known as the NDAC. Over the course of 13 See Smith (1959); Craven and Cate (1955) esp. Ch 8 and 9; Holley (1964); and Fairchild and Grossman (1959), esp. Ch. VI. 14 Higgs (1993) notes that, between July 1, 1939 and June 30, 1940, 87 percent of the War Department procurement occurred through the advertising/invitation-to-bid procedure whereas between July 1, 1940 and February 28, 1941, 74 percent of purchasing was under negotiated contracts. 8 the war, the NDAC begat the Office of Production Management (OPM) which begat the War Production Board (WPB) which begat the Civilian Production Administration (CPA). As was typical in the Roosevelt administration, an additional layer of bureaucracy, first the Supply Priorities and Allocations Board (SPAB) and later the Office of War Mobilization (OWM), was imposed on top of these agencies. Although the agency names changed, the leading actors did not. These included William S. Knudson, a dollar-a-year man on leave from General Motors, Donald M. Nelson, another dollar-a-year man who had been an executive at Sears-Roebuck, and Sidney Hillman, a former union chief. Other principals were Henry Stimson and Frank Knox, two promenient Republicans that Roosevelt had appointed Secretaries of War and Navy, respectively, in the summer of 1940. Most histories of the agencies and officials involved in procurement note that the spending process, especially plant location decisions, induced intense lobbying by politicians and business and community leaders. For example, Nelson, who headed the OPM plant location efforts in 1941, observed: “We were operating in a democracy which was still at peace and subject to the pressures of politics. Platoons of Senators and Representatives stimulated by their constituents, descended upon us. Hundreds of briefs were submitted by towns all over the United States, and, since we were thinking about defense only, I suppose that our selection of sites pleased nobody." Nelson (1946) pp. 149-51.15 The OPM’s official history also notes that during that year the Office “was deluged with requests from Congressman and Senators from various parts of the country suggesting the location of defense plants in their respective Districts and States.” The publication makes special mention of the efforts of Senator Arthur Capper, Republican of 15 According to Nelson's postwar account, the OPM’s Plant Site Committee, which reviewed and approved proposals for new defense plants and facilities, was "instructed to decentralize defense industries, in the interest of employment and raw materials... Naturally, every section of the country wanted plants, but the tendency of the Army and Navy was to place them in areas where the various materials and products had been created before." Nelson, recognizing that this policy was expedient in the short-run, questioned its longer-run effects. "We felt that a very serious manpower shortage might develop... and we thought it important to select locations for new manufacturing facilities in areas where the nation's resources in manpower, transportation and raw materials could be used to best advantage. For instance, there were some sections of the country, notably in the South and Middle West, where large pools of unemployed men existed because there was no sufficient industry to absorb the available labor supply." See also US Civilian Production Administration (1945) pp. 41-42, 56-62. 9 Kansas, to lobby for more mid-western plants.16 Irving Holley’s Buying Aircraft echoes this account. The Army Air Corps procurement officials “responsible for site selection were subjected to a good deal of pressure from various localities in the interior urging their advantages and the need for an equitable distribution of defense orders." In particular, Holley (pp. 307-08) cites letters from Sen. Josh Lee, Democrat of Oklahoma (10 Dec 1940) and Sen. Sheppard, Democrat of Texas (12 Dec 1940) seeking aircraft plants for their states. Frederic C. Lane’s Ships for Victory tells much the same story regarding the U.S. Maritime Commission’s shipbuilding program. The location of new sites was “a question full of political dynamite. Each region was optimistically conscious of its possibilities and would feel slighted if not given what it considered its share. Within each region many more sites were advocated than could be used. Later, a flood of letters came from Senators, Representatives and other political leaders, urging consideration of proposals from their constituents. But the initial selection, which laid down the lines followed in future growth, was made when the political pressure was less than it became after the program was announced and its potentialities were appreciated." p. 47 In a chapter entitled “Politics and Administrative Methods in the Selection of Shipyards Sites,” Lane observes “The whole process of selecting and rejecting was done under heavy political pressure, which was most evident in some cases of rejection…As it became clear that new shipyards were to be financed by the government a torrent of letter poured in to the Commission offering sites. (p. 150)” “(L)abor leaders as well as senators, governors, mayors, and Congressmen voiced local desires to have the government spend money for a shipyard in their neighborhood.(p. 151.)” "Having to reject so many applications, the Commission was exposed to the charge of playing favorites. Individuals who were sent away empty handed felt they or their communities were being discriminated against. There was talk about not having the right ‘contacts.’ (p. 152).” 16 In addition to Capper‘s letter of 29 April 1942, US Civilian Production Administration (1945) p. 57 cites specific letters from representatives from Wisconsin, Arkansas, Louisiana, Montana, Kansas, Indiana, and Connecticut: A. Wiley, 1-29-41; T. Wasielewiki, 1-9-41; C. Ellis, 2-1-41; J. Sanders, 1-15-41; B. Wheeler, 11-26-41; W, Burke, 5-9-41; J. Boehne, 12-23-41; W. Fitzgerald, 3-21-41 10 The placement authorities responded to these complaints by creating Plant or Site Location boards. This counter-move of addressing the problem by adding more bureaucracy is clear in the case of the Maritime Commission. Criticism of its site selection process received a full airing in the hearings of the Truman Committee on 3 June and 9 July 9 1941. (Lane pp. 152-54.) Within a few weeks, the Commission established Shipyard Site Planning Committee to “determine the suitability of projects from the standpoint of geographical position, availability of labor, power, and transportation, and the financial and technical experience of the applicants.” The OPM responded to the political heat generated in plant location process even earlier than the Maritime Commission. In early 1941 “a movement arose in Congress to establish by legislative action a Plant Site Board to pass upon the location of plant sites for Government defense facilities in order to bring about a greater decentralization of industry (US Civilian Production Administration (1945) p. 40).” Noting the “disadvantages of Congress rigidly fixing standards,” William Knudsen suggested the OPM take preemptive action. On 17 March, the Office established a Plant Site Committee “to review and approve or disapprove proposed locations for additional plant or facilities required for the national defense.” The Committee, which was converted into a more permanent Board (or PSB) on 6 May 1941, was to work in close cooperation with representatives of Ordnance Department, the Army Air Corps, and the Navy Department (pp. 40-42).17 “Such factors as availability of labor, transportation facilities, housing, waterpower, community services and attitude, sources of raw materials and destination of the finished products, and the general relation of the new plants to the over-all distribution of manufacturing facilities in the country were carefully examined. The board was anxious to avoid, if possible, the building of plants in already highly industrialized and congested areas.” (p. 56) “The Plant Site Board did endeavor to locate new facilities away from highly industrialized areas. In part the location of new facilities was determined by strategic reasons... According to Nelson, supply contracts followed 17 The PSB continued to operate under the War Production Board. US Civilian Production Administration (1945) p. 107. It was superseded on Oct. 17, 1942 by the Facilities Clearance Board and Facility Review Committee, which continued through May 24 1943. Nelson’s policy was to restrict new construction to facilities that would be on-line before mid-1943. 11 the location of industry; but new facilities were planned to follow at least partial decentralization.” (p. 58.)18 PSB policy called for preserving “the area north of the Mason-Dixon line and east of the Mississippi River for defense manufacturing requiring highly skilled labor, such as aircraft engines, and indicating that approval for other types of facilities in this area would, in general, be given only in exceptional circumstances.” The Board (pp. 60-61) “was aware on the undesirability of further concentrating aircraft facilities in southern California, of expanding plant facilities in the Detroit area, of enlarging shipbuilding plants around Camden, New Jersey, and of locating more plants at Bendix, Philadelphia, Rochester, and other highly industrialized centers.” It acted primarily as a “negative planning unit” which frequently initially vetoed proposed sites and urged the procurement officials look in less congested areas. "In view of the urgency for speeding up production, however, the Plant Site Board was reluctant to exercise this (veto) power for fear of impeding the defense effort.”(pp. 59-61) It appears expediency was the order of the day. The PSB and other civilian authorities generally allowed the military procurement officers to contract where they pleased, and in turn, the procurement authorities allowed their manufacturing suppliers to produce and invest where they saw fit.19 While most official histories cite isolated 18 Donald Davenport, Chief of the Employment and Occupational Outlook Branch of the US Bureau of Labor Statistics, re-iterated this view in early 1942: "It is obvious that both prime contracts and subcontracts have tended to be distributed in accordance with existing production facilities. Contracts for new industrial facilities, however, have been distributed in such a way as to bring about increased geographical dispersion." US Congress, House, 77th Cong., 2nd Sess., Select Committee Investigating National Defense Migration, Hearings, Pt. 27 Manpower of the Nation in War Production: Book One (Washington, DC: GPO, 1942) p. 10258. 19 Holley notes “Even when the government footed the bill as it did during the fall of 1940, responsible air arm officers were unwilling to ignore the contention of many manufacturers that to build secondary plants in the interior, at a distance from parent plants, would seriously slow down production.” Pp. 307-08. According to US Employment Service (1948), much the same can be said about civilian labor authorities. During the NDAC and OPM periods, the US Employment Service supplied labor market data to the OPM's Plant Site Board, but it was used only to 'a limited degree ... in locating new facilities...In a period of large labor surpluses, manpower implications were of limited importance." Pp. 114. In August 1942, Facilities Clearance Board took over, providing WMC a more formal advisory role in the location of government offices. Supply contracts were still awarded by procurement officers in numerous branches. WPB Directive No. 2 of March 1942 established procurement policies, including provisions (added at the behest of the WMC and WPB) that agencies take account of critical labor supply conditions. As amended in April 1943, it enjoined procurement agencies "to avoid contracting for production of items or materials in communities or areas in which labor shortages are known to exist, whenever it is practical to procure the needed items or materials elsewhere." After 1942, the USES and later the WMC issued monthly reports designating areas with (I) current labor shortages (II) anticipated labor shortages, and (III) abundant labor 12 instances of successful application of political pressure or of questionable plant/contract placement, they generally paint a picture of spending decisions that were economically and strategically motivated. The evidence is clear that politics or peacetime policy objectives played crucial roles in some important wartime spending decisions. Dating back to his service in the Department of the Navy during World War One, Roosevelt had “been enthusiastic for shipyards in the South.” Under the chairmanship of Joseph Kennedy in 1938, the US Maritime Commission received congressional permission to grant contracts to shipyards in the South and West despite their higher cost structures (Lane (1951), pp. 102-04). Although the performance of southern shipbuilders remained below eastern levels in the early 1940s, the Commission followed the administration’s wishes by granting some wartime contracts to southern yards. Costs and productivity on the West Coast did reach parity with the east by the early 1940s, leading to the placement of large share of contracts there during the war. But the pre-war West possessed no modern integrated steel plants and hence no capacity to produce ship plates locally. In response, Roosevelt had the federal government help finance two new steel plants (at Geneva UT and Fontana, CA) in the West, satisfying another of his long-term policy objectives. In addition, there were numerous accusations of influence peddling, kickbacks, and conflicts of interest regarding defense spending. Notable contracting scandals involved Thomas Corcoran, a New Deal political operative, General Bennett Meyers of the Army Air Crop, Representative Andrew May of Kentucky, chair of the House Committee on Military Affairs, and Senator Theodore Bilbo of Mississippi.20 Finally, there is no mistaking that the war era was a period of political flux, as the evidence in Table 4 indicates. On the surface, the Democrats appear uncontestable “in charge” with the Presidency and large majorities in both House of Congress as the war broke out. But FDR’s relatively easy victories over Wendell Willkie in 1940 and Thomas Dewey in 1944 should not blind us. (Notice that in each election, the margin of even at the war production peak. Headquarters officers accepted the principle of preferring group (III) areas to group (I) but officers in the field did not apply these rules. p. 116 As amended Oct. 10, (1942?), the WMC issued rules that contracts could not be given to Group I areas if other facilities were available; Supply contracts of longer duration than 6 months could not be given to group II areas. Contracting in group III areas was unrestricted. 20 US Congress, Senate, 76th Cong., 2nd Sess., Special Committee to Investigate the National Defense Programs, Hearings (Washington, DC: GPO, 1941-47). 13 victory was smaller than before.21) Roosevelt’s success appears in large part a product of his own personal popularity and of the troubled times, rather than the health of his party. The absence of a creditable successor during a period of international crisis motivated Roosevelt’s decision to seek his unprecedented third and fourth terms. The Democrat’s 1940 campaign slogan “Don’t Change Horses in Mid-Stream” apparently captured the mood of many in the electorate. But it also hints that many voters had a desire to change horses, which they might exercise once the stream was forded. Indeed, the experience after World War One suggests that “the return to normalcy” was associated with a change in the party in control.22 But the real sign of the extent of political flux during the war period appears in Congress. In the 1942 midterm elections, the Republicans picked up 46 seats in the US House of Representatives, reducing the Democrat’s majority from 106 to just 10 seats. Even more striking is that collectively the Republican congressional candidates outpolled the Democrats for the first time since 1930. In the Senate, the Republicans picked up 9 seats in 1942, winning 18 out of the 36 races contested. The Republicans continued to pick up Senate seats in the 1944 elections although they backtracked slightly in the House. And in the 1946 midterm elections, the Republicans won majorities in both House. Given the state of political flux, the allocation of money to improve one’s political prospects seems entirely plausible. III How important then were political motivations compared with economic and strategic considerations in determining contracting and investment decisions during World War Two? The literature on New Deal spending offers a promising approach to address this question, but it is desirable first to extend the model. This section develops a 21 Of course, it would have been difficult for FDR to top his1936 landslide when he won 523 electoral college votes to Alfred Landon’s 8. The magnitude of this victory margin raises some questions about the relevance on New Deal vote buying, especially in the late 1930s. 22 Political motivations could remain relevant for explaining spending even if a change in government was guaranteed. Instead of using its spending authority to buy swing voters, the outgoing party could opt to reward its base. 14 simple model of vote-buying that goes beyond the existing approach by encompassing (1) the allocation of spending across jurisdictions within an election area; and (2) the joint role of Presidential and congressional candidates in determining spending outcomes. The existing historical literature (Wright (1974), Wallis (1987, 1998, 2001)) focuses essentially on the allocation of spending across states to win the Presidency. The central concept is the state’s political productivity, which is measured by the its electoral college votes times the change in the probability of winning the state given the historical pattern of the state’s election outcomes. The implications are that spending is relatively more productive in states with high electoral votes per capita, with mean vote shares close to one-half, and with higher variances.23 The implications of the state-level model do not carry over directly to explaining the allocation of spending within states. Most obviously, the absolute difference of a county’s vote from 50 percent has no bearing on the desirably of spending there. Given the variations in the electoral composition, a candidate may be able to spend in areas that both rewards loyal supporters and buys swing votes at the same time. This is not a possibility in the state-level models. More importantly, simple theoretical considerations reveal that the most relevant measure for a county is not its vote share, but rather its net vote margin. To develop these insights further and provide better grounding for our empirical work, we now build a simple model of vote buying at the sub-unit level in winner-take-all elections.24 We begin by assuming a politician of the incumbent party, say the Democrats, has a total of Y dollars to spend in his election area. He wishes to allocate these funds across the local jurisdictions to maximize his probability of winning the next election. Assume each jurisdiction i has a voting population, Ni, composed of three types of voters—loyal Democrats numbering NDi; loyal Republicans numbering NRi; and swing 23 The role of variability has been given two interpretations. In the first, states with high variation have a smaller density at one-half and in Wallis’s words (1998, p. 148), “require greater expenditures in order to ensure a majority with a given level of confidence.” In the second interpretation, variable voters are easier to buy with spending. 24 The model is similar to Lindbeck and Weibull (1993) who also consider competition for swing voters. The main difference is that they consider a game between two parties for a single election while we consider a single actor who is contesting several elections. Also, the control variable in Lindbeck and Weibull is a single policy over which voters have divergent opinions while our actor allocates a local public good over the different election areas. 15 voters numbering NSi. Thus Ni≡NDi+NRi+NSi. (Note we are assuming each type of voter behaves the same across all of the jurisdictions in the election area. The jurisdictions differ solely in the composition of the electorate across the three groups.) Assume the loyal Democrats and Republicans always turnout to vote for their candidates. Assume that the fraction of swing voters in jurisdiction i casting a Democratic ballot depends on random shock, p, and on per capita spending in the jurisdiction, yi≡Yi/Ni. 25 Specifically, the fraction equals θp+(1-θ)(1+(F(yi))/2) where θ∈[0, 1]; p has a continuous, differentiable distribution G(x) with G(0)=0, G(1)=1 and density g(x)≡G’(x)>0; and F(y) is a continuous, concave function such that F’(y)≡f(y)>0, F”(y)<0, F(0)=0, and F(∞)=1. Notice that in the absence of spending, the swing voters favor the Democrats if p>1/2. To simplify notation, let y be the vector of the yi in the election area. Given yi and p, the total number of Democratic and Republican votes in jurisdiction i are, respectively: VDi =NDi+NSi (θp+(1-θ)(1+F(yi))/2) VRi =NRi+NSi (θ (1-p)+(1-θ)(1-F(yi))/2). The net margin is: VDi-VRi =NDi-NRi +NSi ((2p-1)θ+(1-θ)F(yi)). Notice that the variance in votes in the jurisdiction, 4θ2NSi 2Var(p), is increasing in the number of swing voters. Denote NX≡ΣiNXi for X=D, R, or S, and VX≡ΣiVXi for X=D or R. Summing up the net margins across jurisdictions, the total net margin is VD-VR=Σi(VDi-VRi) =Σi(NDi-NRi+NSi((2p-1) θ+(1-θ)F(yi))). 25 The underlying idea behind this per-capita formulation is that though the politician can determine the number of swing voters in each jurisdiction in his district, he cannot target individual swing voters. Otherwise, he would just buy the cheapest vote he could find. 16 Given the allocation of spending, y, the probability of a Democrat win is: Prob(VD>VR|y)= Prob[p>1/2-((ND-NR+(1-θ)ΣiNSiF(yi))/(2θNS))|y]  0 if (1-θ)ΣiNSiF(yi)+ θNSVR|y) subject to Y≥ΣjYj. Ignoring corners in (1), the objective function may be written as Prob(VD>VR|y) = 1 – G(Ω(y)). Since G(x) is monotonically increasing, argmaxy 1 – G(Ω(y)) = argmin y Ω(y). Hence, we will solve the transformed problem: (3) Minimize Ω(y) subject to Y≥ΣjYj. Notice that Ω(y) is convex and decreasing in y. The first order condition of (3) states that in all jurisdictions, k, with NSk>0 funds should be allocated according to: (4) f(Yk/Nk)(NSk/Nk)(1-θ)/(2θNS)= ψ. Notice that the Lagrange multiplier, ψ, from the transformed problem (3) can be related to the multiplier, λ, for the original problem (2) according to λ≡ψg(Ω(y)). Equation (4) means that under the optimal allocation, the value of buying a vote in each jurisdiction is equal. A further interpretation is that at the optimum, 26 Note if ND>NR+NSθ, the Democrat is guaranteed to win. Similarly if NR>ND+NS, a Republican win is assured. There is an asymmetry here because we are assuming the incumbent, who has the ability to buy votes, is a Democrat. 17 (5) ψg(Ω(y)) ≡ λ = ∂Prob(VD>VR)/∂Y. λ has the standard interpretation of the marginal value of an extra dollar of total spending. Inverting (4) yields a formula for optimal spending in k. (6) Yk=Nk f-1(2ψθNkNS/(NSk(1-θ))). Note because by the concavity of F(y), df-1(z)/dz<0. Thus as the fraction of the electorate comprised of swing voters (NSk/Nk) increases, per capita spending increases. Note a corner solution, with Yk =0, is possible under two circumstances. First, there will be no spending in a jurisdiction with no swing votes, NSk=0. Second, in election areas which are non-contested, (that is, with either NR>NS+ND and the Republicans win with probability one or ND>NR+NSθ and the Democrats win), then any allocation of funds is optimal.27 Notice that the composition of the electorate in an individual jurisdiction between Democrats and Republicans has no effect, recreating a useful exclusion restriction in the empirical model. This model readily extended to two-stage, winner-take-all contests such as gaining the Presidency or control of the House or Senate. In such contests, one must win individual elections at the district or state level. The contests may differ in their importance through weights such as the number of votes in the Electoral College. Consider a model where the Democratic candidate for president seeks to maximizes the expected number of Electoral College votes by winning the state-level contests: (7) Maximize ΣCΚCProb(VDC>VRC|y) subject to Y≥ΣCΣjYjC where ΚC is the Electoral College vote of state C and YjC is the spending in jurisdiction j in state C. 28 In the spending allocation problem, the candidate uses a two-stage solution. In the first stage, the candidate chooses the level of funds, YC, to allocate to the state C. 27 Once (1-è)ΣiNSiF(yj)+èNS=ND-NR, a corner-solution again results because effectively ë=0. 28 Note this model differs slightly for the actual Presidential contest where the candidate seeks to win a majority of the Electoral College votes. But it seems justified by the advantages in claiming a mandate to govern by gaining more than a bare majority of the votes. 18 In the second stage, the candidate optimally allocates the budgeted funds across the jurisdictions in the state to maximize the probability of success. Using YC≥ΣjYjC in the solution approach discussed above solves the second-stage problem. Now the solution to the first stage problem associated with (7) is, (8) ΚC ∂Prob(VDC>VRC)/∂YC=µ where µ is the Lagrange multiplier on the budget, Y≥ΣCYC. This implicitly determines the distribution of spending across states, YC. Applying the solution to the second stage given in (4) and (5), this becomes, (9) YjC=NjCf-1(2µθNjCNSC/(ΚCNSjC(1-è)g(Ω C(yC)))), Presuming that each jurisdiction is small (implying YjC has a negligible effect on the density g(Ω C(yC)), equation (9) is the formula for optimal spending. It states that spending in a jurisdiction is influenced by state-level variables- -increasing in Electoral College votes and the density of the conditional outcome, and decreasing in swing votes state-wide – and by its own local variables--increasing in the local swing votes.29 29 A straightforward extension allows this two-stage approach to model the problem of winning or maintaining power in Congress. Congressional districts replace states as the election area and weights (reflecting leadership positions and seniority) may replace the Electoral College votes. One difference is that the Democratic congressional leadership will not want to spend in Republican districts because presumably the Republican incumbent will be able to claim credit for this spending. The net margin equation will be modified to become VD-VR=Σi(VDi-VRi) =Σi(NDi-NRi+NSi((2p-1)è-(1-è)F(yi))). The model can be further extended to capture concurrent or overlapping elections, for example, with the President (or Senator) and Representatives running at the same time. Consider the simple example where the R congressional candidates and a presidential candidate are running within a single state. (We will here treat the congressional district as the smallest unit of analysis.) Let vote buying be characterized by joint production with candidates at both levels receiving credit for the spending. More specifically, assume that voters in each district vote a straight ticket, so the congressional and presidential candidates piggy-back perfectly. Further assume the Party assigns a weight ðr to congressional candidate r winning his contest and a weight (1-Σrðr) to the presidential candidate winning the state as a whole. Then the Party’s problem is: Maximize (1-Σrðr) Prob(ΣrVDr>ΣrVRr|y) +ΣrðrProb(VDr>VRr|y) subject to Y≥ΣrYr. Ignoring non-contested elections, the first-order condition is (ðr+(1-Σrðr)(NSr/NS))(f(Yr/Nr)/Nr)(1-è)/2è =ë. The key observation is the presidential candidate cares only about the net margin aggregated across the congressional districts whereas each congressional candidates care only about the net margin in his own district. This will lead to a distortion in spending from the pattern in a President-only contest. Specifically, from the presidential candidate’s perspective, spending will be shifted away from safe districts with cheap swing voters to more contested districts with more expensive swing voters. The extent of the distortion will be greater the higher are the ðr. More powerful representatives, those with a relatively high individual ðr in the party’s preferences, will be allocated more funds all else equal. This provides a way of testing “whose’s in charge?”—Congress or the President. 19 IV In our empirical analysis, we investigate the relative importance of political determinants in the geographic distribution of World War Two spending. Our county- level data for military spending, covering the period from June 1940 to June 1945, include total defense spending per capita as well as per capita contracting and facility investment, separately. The distribution of spending at the county level displays two important statistical differences from spending at the state level. First, as noted above, there are numerous “zeros” in the county-level data whereas every state receives positive spending. This requires using different econometric techniques than much of the previous literature. Second, the spending distribution across counties receiving positive levels was highly skewed. Working with per capita spending in logs fits the data better that working with it in linear terms (as Wright (1974) and Wallis (1987, 1998, 2001) did in their studies of the state-level distribution of New Deal spending.) As a first step in the econometric analysis, it is necessary to modify the vote- buying model of the previous section into a form that can be empirically implemented. The key result of that model was equation (9). Cross-multiplying by NC, the jurisdiction’s size, equation (9) can be re-stated as, (10) YjC/NjC ∝ gC×(ΚC/NC)(NsC/NC)-1(NsjC/NjC) where the proportionality is used because f-1(.) and various constants are omitted, and gC is a term which reflects the density of the shock in election area C. Two more steps are needed to take (10) to the data. First, recalling the variance relationships and presuming that vote turnout is proportionate to population, NsjC/NjC ∝ Standard Deviation(DVSjC) (11) NsC/NC ∝ Standard Deviation(DVSC) 20 where DVS≡VD/N is the Democrat’s vote share. Second, maintaining the presumption that each jurisdiction is small (so gC is independent of YjC) and adding the assumption that the density of the shock is a (weakly) single-peaked and symmetric about 0.5, (12) gC ∝ -|0.5-DVSC| The right-hand sides of (11) and (12) can be approximated using historical election data (presuming that there was no earlier allocation of funds). Combining (10)-(12), this suggests the regression, (13) Ln(YjC/NjC) = α1×|0.5-DVSC| + α2×(ΚC/NC) + α3×Standard Deviation(DVSC) + α4×Standard Deviation(DVSjC) Theory predicts that α1<0, α2>0, α3<0, α4>0. In words, this regression considers how the allocation of resources to a jurisdiction are influenced by its electoral responsiveness to funding as well as the contestability, political importance and electoral responsiveness of its election area. In practice, one modification must be made to (13). Because there may be fixed costs to allocating resources to a jurisdiction (such as creating a local monitoring institution) and because small contracts are omitted from the data, jurisdictions may receive a zero allocation even if their election area is not at a corner solution. To the extent that certain variables influence whether any funds are allocated but not the conditional funding level (such as land area), this suggests (13) should be estimated in two parts. We will estimate separate equations for whether a jurisdiction gets funding, and also an equation for the conditional funding intensity. Various estimators of the equations, such as the two-step model and a Heckman sample selection model which allows correlation of the errors, will be considered. In the above, we have typically interpreted KC as the number of electoral votes. But note that in a congressional model it can be readily re-interpreted as the individual Senators or Representatives weight in the decision-maker’s objective function. In our empirical implication, we can capture this effect by including variables reflecting the politician’s position in the congressional power structure such as party, seniority, 21 leadership position, and membership on the key committees making the allocation decisions.30 In addition, we can include a variable on the number of representatives per capita (analogous to the number of electoral college votes per capita) to capture the vast inequalities in representation prevailing in the era before the one-man one-vote decision. To provide a full test of the role of politics, it is important to develop a baseline model capturing the procurement authorities’ stated economic and strategic objectives. These included utilizing existing capacity first, decentralizing production away from congested areas with tight labor markets, and locating production in the nation’s interior to avoid enemy attack and espionage. To capture these economic objectives, we include two set of variables: (a) measures of under-utilized resources, including the fraction of the labor force which was unemployed in 1940 and the fraction of the county's population in rural areas; and (b) measures of the county's pre-war manufacturing capacity, including the number of wage-earners per capita in 1939 and the number of manufacturing establishments in aircraft, automobile, ordnance, and shipbuilding industries in the county in 1935.31 (To capture potential spillover effects across county lines, we include measures of manufacturing capacity and population in neighboring counties. A neighboring county was defined as one with its seat of government within forty miles of subject county’s seat.) To capture strategic considerations, we include dummies for whether the county was on the coast or interior. Strategic doctrine, dating to 1915, held that munitions contracts should be placed at least 200 miles from the coastline and the borders with Mexico and Canada. We also include variables for the pre-war military capacity, specifically dummy variables reflecting the presence of an army or navy base in the county in 1937. Most of these bases date to the World War One or before, implying that problems of endogeneity with 1940-era political considerations are likely to be unimportant. (Again to capture spillover effects, we include dummies for the presence of such bases in neighboring counties). 30 The most relevant committees were Appropriations, Military Affairs, Naval Affairs, Maritime (House only.) 31 The automobile industry combines “motor vehicles” and “motor vehicle bodies and parts.” The ordnance industry combines “firearms” and “ammunition.” The year 1935 was chosen both because establishment data were available at the county level (unfortunately due to census disclosure rules, employment and output are not) and to avoid issues of endogeneity whereby the politics of the immediate pre-war period influence the location of these industries. 22 Table 5 provides the summary statistics for the variables used in the analysis. The Data Appendix contains full details on the data definitions and sources. V Table 6 reports the results of our Heckman Maximum Likelihood estimates of the determinants of total spending, facility investment, and contract spending. The Table presents two sets of results for each equation—one using presidential election variables and the second using congressional variables. (Multicolinearity among the election variables complicates efforts to include both sets of variables in a single equation.) We will begin our analysis focusing on the presidential equations (1)-(3). The results for total spending and its subcomponents--facility investment, and contract spending—differ in many details.32 But a common pattern emerges: money tends allocated to urban counties, those on the coast, and those with pre-existing military or manufacturing capacity. In particular, pre-war military bases were highly important for facility investments. And proximity to pre-war manufacturing capacity (both in terms of specific industries such as shipyards and auto plants and in terms of the density of manufacturing wage-earners) mattered significantly for contract spending. The coefficient of the dummy for shipyards is especially notable. Manufacturing capacity did not have a significant effect on facility investment except for aircraft plants in neighboring counties (and manufacturing wage-earners in the selection equation).33 Strategic considerations such as seeking secure areas in the country’s interior appear to receive relatively little weight. Comparing the dummy variables for coast and secure reveals that counties on the coast received greater per capita spending than those is areas 200 miles from the nation’s border or coastline. (Secure counties did receive more money than non-coastal, non-secure counties.) Objectives such as seeking pockets of 32 Using a likelihood ratio test, we can reject at even the 99 percent confidence level the hypothesis that common coefficients in the facility investment and contracts equations. This implies that the two types of spending responded differently to the explanatory variables. 33 Note that in the selection equation that every location possessing either an ordnance or aircraft plant in 1935 receives positive contract spending. These observations are dropped in the selection equation in the two-step Heckman mode. The presence of an ordnance or aircraft plant in a neighboring county has a significant, positive effect on the probability of receiving funding.) 23 underemployed labor also were unimportant. Spending was strongly negatively correlated with the rural variable in all of the regressions and was unrelated to the county’s 1940 unemployment rate. To control for spillover effects between neighboring counties, we included variables for the total population and number of manufacturing wage-earners in the neighboring counties relative to the subject county’s population. A higher number of wage-earners nearby tended to increase the county’s total and contract spending whereas a larger neighboring population tended to reduce spending in these categories. (These spillover variables had little discernable effect on facility investment.) Variables capturing the power of the county’s congressional members generally lack explanatory power either in the spending equation or in the selection equation.34 In particular, these findings do not support the view that senior members of the majority party brought home the pork. Actually having a senior senator who is a Republican tends to measurably increase spending (a Hiram Johnson effect?). And longer tenure in the house was associated with less funding. Membership on the house key committee was associated with higher levels of contract spending. We cannot determine whether the membership effect was due to these representatives actually increasing spending in their districts or whether representatives from military-related districts sought membership on these committees. The overall impression for the congressional power coefficients is that they are remarkably small. The presidential election variables bear a more interesting relationship to spending. In general, there is a U-shaped relationship between Electoral College votes per capita and spend money.35 Counties in states with a high number of Electoral College votes per capita (extremely small states such as Nevada) and those with low number (populous states such as New York) receive more spending the intermediate states. This makes sense in terms of the standard models of winner-take-all elections. In these models, the value of big states is disproportionately large because their Electoral College delegations are much more likely to be pivotal than those of smaller states. A small state 34 These results are consistent with Gist and Hill (1984) who find that committee assignments do not have a significant effect on the geographic allocation of funds which are nominally controlled by HUD bureaucrats. 35 Note the log of electoral college votes per capita is negative and its square positive. Raising the number of electoral votes per capita increases the log term towards zero and reduces its square. Note the electoral votes per capita variables appear to have little effect in the selection equation. 24 (with a high ratio of Electoral College votes per person) can offset this disadvantage by being so cheap to buy. The swing voter variables do conform to the theory in the total spending and especially in the facility investments equations. Counties in states where the democratic presidential vote share is closer to 50 percent receive more money. Consistent with the swing-voter story, greater variability of democrat share at the county-level has a significant positive effect on facility investment in the county. Greater variability at the state level has a negative effect, consistent with the interpretation that swing voters are cheaper elsewhere. Finally, loyalty mattered. Counties with traditionally high democratic share received more facilities investment. (This effect does not disappear if a dummy for the South is included.) Note that using county-level allows us to capture this effect – spending where both the candidate’s base is strong and the number of swing voters in high—which would be masked in state level analysis.36 Turning to the congressional election models, we have replaced presidential variables such as electoral votes per capita with the number of representatives per capita and the historical presidential elections with their congressional analogues.37 Do these changes make any difference? The fit of the facility equation is slightly better than before, that of the contract spending equation worse. The representation variables behave much as the Electoral College variables did. Spending has a U-shaped pattern with areas with a high number of representatives per capita and those with a low number receiving more spending of each type than those with an intermediate number. This pattern contrasts with results reported in the literature for the recent period where spending rises continuously with representatives per capita.38 (Under today’s one-man one-vote regime, high representation results principally from small state population.) The results for the swing-voter story for the congressional election are more mixed. The “closeness” of the election does not have a significant impact as in the presidential models. The other election variables have similar signs to those in the presidential equation, but the significance levels are lower. Note loyalty to the democrat party still pays. The 36 Basically in state-level analysis, a strong base would mean the election will not be closely contested and buying votes is unnecessary. 37 We have made no attempt to include Senator election results to this point. 38 See Atlas et al (1995) who find that greater per capita Congressional representation, particularly in the Senate, increase federal per capita spending in a state. 25 coefficient of the democratic share in the facilities investment equation (and as a consequence in the total spending equation) is still strongly positive. Politics matter here, but mainly through rewarding the party faithful rather than buying the swing votes in close elections. Results in the Two Part Model are shown in Table 7. These estimate the level of spending conditional on receiving funds, ignoring the Heckman sample-selection correction. Notably for our purposes, they are very close to those reported above, providing greater confidence in the robustness of these findings. To provide a sense of the magnitude of these effects, we will now investigate a series of “experiments.” We explore how per capita spending would increase in a county in which key variables were shifted one standard deviation from the mean or for which sets of indicator variables changed from their mean to unity or “on.” Specifically, consider the following six experiments: (1) changing all of the swing-voter variables (the electoral college, election closeness, the voter variability terms) one standard in the direction that the theory suggests raises spending; (2) increasing loyalty to the democratic presidential candidate by one standard deviation; (3) shifting all of the discrete congressional power variables (party affiliation, leadership, and committee membership) in the direction of greater power;39 (4) raising the number of manufacturing wage-earners per capita in the county (and its neighbors) by one standard deviation; (5) shifting all the plant indicators from their mean position to “on”; and finally (6) turning all of the military installations indicators “on.” In all cases, we will use means and standard deviations from the entire population of counties (not just those receiving funds). The baseline value for total spending per capita, derived by taking the exponent of the product of the population means times the coefficients in the presidential equation, is $93. The difference from the national average arises in part from using calculating county means without population weights and from Jensen’s inequality (which implies the expectation of the log of a variable differs from the log of its expectation.) See Manning and Mullahy (2001) for a discussion of the difficulties of interpreting log models. 39 To avoid mixing standard deviation changes and turning “on” indicator variables, we ignore the effects of increasing tenure, which are negative in any case. 26 The results of these experiments for the presidential total spending are presented in Table 8. Table 8: Simulated Effects Experiment Spending Change 0. Baseline $ 93 1. Swing Voter $ 134 $ 41 2. Democratic Loyalty $ 108 $ 15 3. Congressional Power $ 78 $ (15) 4. Mfg Wage-Earners $ 246 $ 153 5. War Mfg Plants $ 611 $ 518 6. Military Bases $ 380 $ 287 As indicated in experiment 1, the swing-voter variables collectively have an economically meaningful effect on spending—when each is changed one standard deviation in the relevant direction total spending per capita increases by $40. Loyalty also pays—increasing the county’s democratic vote share by one standard deviation (that is, 19 percent) above the mean (54 percent) raises spending by $15. Congressional power has little economically meaningful impact. By contrast increasing industrial capacity (measuring either by raising the per capita manufacturing wage-earners variables or by turning the plant dummies “on”) has dramatic effects. Increasing the wage-earner variables by one standard deviation raises spending by $153 per capita. Shifting all of the plant variables from their mean positions to “on” increased total spending by $518 dollars. In the final experiment when all of the military base variables are shifted from their means to “on”—roughly the equivalent of the change from the typical county to say Norfolk Virginia, leads to the greatest increase in spending, by $287. The take-home message of this analysis is that politics mattered, especially in the allocation of facility investments, but that the major determinant of the geographic allocation of World War Two spending was the region’s pre-1940 military and manufacturing capabilities. VI 27 This paper assembles and analyzes a new county-level dataset including economic and political variables to gain a deeper understanding of the geographic distribution of World War Two military spending. Our results indicate that political factors such as electoral importance, party loyalty, and the cost of votes shaped spending decisions on the margin. Congressional power, however, had little apparent impact. In fact, the overwhelming determinant of spending was the pre-existing military installations and manufacturing capacity. The importance of the economic variables in the World War Two period contrasts with the emphasis in the New Deal literature on the primacy of politics. It also suggests that contemporary analysis of determinants of federal resource allocation should play closer attention to economic factors. Several directions for future research suggest themselves. The first is to conduct a comparable study of the determinants of county-level spending during the New Deal. The second would be to explore how spending during the Second World War affected election outcomes and party affiliation (that is, to estimate a system of equations relating spending and voting). The third and perhaps more fundamental task is to develop more refined measures of the population of swing voters. Relying on the variance of historical outcomes seems less satisfactory than developing a more empirical-based characterization of which demographic groups behaved as swing voters, which shifted between parties, and which were most responsive to economic incentives. More generally, it would be desirable to gain a fuller understanding of the relationship between voting behavior (e.g. partisan turnout) and spending. Finally, future research could develop better measures of the welfare effects of political influence on federal resource allocation. 28 REFERENCES Anderson, Gary M. and Robert D. Tollison (1991). “Congressional Influence and Patterns of New Deal Spending, 1933-1939” Journal of Law and Economics. 34:1 (April): 161-75. Anderson, Paul H. (1948). State, Regional, and Local Market Indicators, 1939-46. Office of Domestic Commerce, Economic Studies No. 67 (Washington, DC: GPO). Atlas, Cary, Thomas Gilligan, Robert Hendershott and Mark Zupan (1995). “Slicing the Federal Government Net Spending Pie: Who Wins, Who Loses and Why.” American Economic Review. 85: 624-629. Bateman, Fred, and Jason Taylor (2001a). “Was “V” for Victory or Votes? A Public Choice Analysis of World War Two Federal Spending.” unpublished working paper. Bateman, Fred, and Jason Taylor (2001b). “The New Deal at War: New Deal Agencies’ Expenditure Patterns, 1940-1945.” unpublished working paper. Clubb, Jerome M., William H. Flanigan, and Nancy H. Zingale (1986). Electoral Data for Counties in the United States: Presidential and Congressional Races, 1840-1972 [Computer file]. ICPSR ed. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [producer and distributor]. ICPSR 8611 Connery, Robert (1951). The Navy and the Industrial Mobilization in World War II (Princeton, NJ: Princeton Univ. Press). Craven, Wesley F. and James L. Cate (1955). Army Air Forces in World War II, Vo. 6, Men and Planes (Chicago: Univ. of Chicago Press). Fairchild, Byron and Jonathan Grossman (1959). Army and Industrial Manpower. (Washington, DC: GPO). Fishback, Price V., William C. Horrace, Shawn Kantor, (2001). “The Impact of New Deal Expenditures on Local Economic Activity: An Examination of Retail Sales, 1929- 1939” NBER W8108, Feb. 2001. Fleck, Robert K. (1999). “Electoral Incentives, Public Policy, and the New Deal Realignment.” Southern Economic Journal 65, 377-404. Fleck, Robert K. (2001). “Population, Land, Economic Conditions, and the Allocation of New Deal Spending.” Exploration in Economic History, 38:2 (April): 296-304. 29 Gist, J. and R. Hill (1984). "Political and Economic Influences on the Bureaucratic Allocation of Federal Funds: The Case of Urban Development Action Grants." Journal of Urban Economics. 16: 158-72. Goss, Carol F. (1972.) “Military Committee Membership and Defense-Related Benefits in the House of Representatives.” Western Political Quarterly 25: 215-233 Groves, Leslie R. (1962). Now It Can Be Told: The Story of the Manhattan Project (New York: Harper and Brothers). Hewlett, Richard G. and Oscar E. Anderson, Jr. (1962). The New World: A History of the United States Atomic Energy Commission, Vol. 1, 1939/1946. (Oak Ridge, TN: US AEC Technical Information Center). Higgs, Robert. (1993). "Private Profit, Public Risk: Institutional Antecedents of the Modern Military Procurement System in the Rearmament Program of 1940-41." in Sinews of War: Essays on the Economic History of World War II, ed. by Geofrey T. Mills and Hugh Rockoff (Ames: Iowa State Univ. Press) pp. 166-198. Holleran, Owen Cobb (1938). Industrial Market Data Handbook of the United States. US Bureau of Foreign and Domestic Commerce, Domestic Commerce Series no. 107 (Washington, D.C. U.S. Dept. of Commerce). Holley, Irving (1964). Buying Aircraft: Materiel Procurement for the Army Ai Forces. (Washington, DC: GPO). Inter-university Consortium for Political and Social Research. United States Historical Election Returns, 1824-1968 [Computer file]. 2nd ICPSR ed. Ann Arbor, MI: Inter- university Consortium for Political and Social Research [producer and distributor], 1999. ICPSR 1. Inter-university Consortium for Political and Social Research. Historical, Demographic, Economic, and Social Data: The United States: 1790-1970 [Computer file]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [producer and distributor], ICPSR 3. Lane, Frederic C. (1951). Ships for Victory: A History of Shipbuilding Under the U.S. Maritime Commission in World War II (Baltimore, The Johns Hopkins Press). Levitt, Steven and James Snyder (1995). “Political Parties and the Distribution of Federal Outlays.” American Journal of Political Science. 39: 958-980. Levitt, Steven and James Snyder (1997). “The Impact of Federal Spending on House Election Outcomes.” Journal of Political Economy. 105: 30-53. Lindbeck, Assar and Jorgen Weibull (1993). “A Model of Political Equilibrium in a Representative Democracy.” Journal of Public Economics. 51: 195-209. 30 Kester, Randall B. (1940). “The War Industries Board, 1917-1918: A Study in Industrial Mobilization” American Political Science Review, 34:4 (Aug.) pp. 655-684. Manning, Willard G. and John Mullahy (2001). “Estimating log models: to transform or not to transform?” Journal Of Health Economics (20):4 (July): 461-94 Mayer, Kenneth R. (1991). The Political Economy of Defense Contracting. New Haven: Yale University Press. Miller, Nelson A. (1947). State, Regional, and Local Market Indicators, 1939-45. Office of Domestic Commerce, Economic Studies No. 60 (Washington, DC: GPO). Nelson, Donald M. (1946). Arsenal of Democracy: The Story of American War Production (New York: Harcourt-Brace). Ray, Bruce A. (1981). “Military Committee Membership in the House of Representatives and the Allocation of Defense Department Outlays.” Western Political Quarterly 34: 222-234. Reading, Don C. (1973). “New Deal Activity and the States, 1933 to 1939” Journal of Economic History 33: 3 (Dec.): 792-810 Rhode, Paul W. (2000) “The Impact of World War Two Spending on the California Economy” in R. Lotchin (ed.), The Way We Really Were: The Golden State in the Second Great War (Urbana: University of Illinois Press) pp. 93-119. Rhodes, Richard (1986). The Making of the Atomic Bomb (New York: Simon and Schuster) Rundquist, Barry S. (1978). “On Testing a Military Industrial Complex Theory.” American Politics Quarterly 6: 29-53. Rundquist, Barry S., and David E. Griffith. (1976). “An Interrupted Time-Series Test of the Distributive Theory of Policy-Making.” Western Political Quarterly 24: 620-626. Smith, R. Elberton (1959). The Army and Economic Mobilization (Washington, DC: GPO). U.S. Bureau of the Census. County and City Data Book [United States] Consolidated File: County Data, 1947-1977 [Computer file]. Washington, DC: U.S. Dept. of Commerce, Bureau of the Census [producer], 1978. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 1981.ICPSR. 7736 U.S. Civilian Production Administration (1945). Bureau of Demobilization. “The Facilities and Construction Program of the War Production Board and Predecessor 31 Agencies, May 1940 to May 1945.” Historical Reports on War Administration: War Production Board, Special Study No. 19 (Mimeo, 1945). U.S. Civilian Production Administration (1946). Bureau of Demobilization. “Chronology of the War Production Board and Predecessor Agencies, August 1939 to November 1945,” Historical Reports on War Administration: War Production Board, Misc. Publ. No. 1 (June 20, 1946). U.S. Civilian Production Administration (1947). Bureau of Demobilization. Industrial Mobilization for War: History of the War Production Board and Predecessor Agencies, 1940-45 Vol. I Program and Administration. Historical Reports on War Administration: War Production Board, General Study No. 1 (Washington, DC: GPO, 1947). U.S. Congress, Official Congressional Directory Official Congressional Directory 77th Cong 2nd Sess, Corrected to Dec. 19. 1941, Washington DC: U.S. G.P.O. U.S. Congress, Official Congressional Directory Official Congressional Directory 78th Cong12nd Sess, Corrected to May 14. 1943, Washington DC: U.S. G.P.O. U.S. Employment Service (1948). A Short History of the War Manpower Commission. Technical Service Division, US Department of Labor, (June). Vander Meulen, Jacob (1991). Politics of Aircraft: Building An American Military Industry (Lawrence KS: Univ. Press of Kansas). Wallis, John J. (1987). “Employment, Politics and Economic Recovery during the Great Depression.” Review of Economics and Statistics 69:3 (August): 516-20. Wallis, John J. (1998). “The Political Economy of New Deal Spending Revisited, Again; With and Without Nevada.” Exploration in Economic History, 35:2 (April): 140-70. Wallis, John J. (2001). “The Political Economy of New Deal Spending, Yet Again: Reply to Fleck” Exploration in Economic History, 38:2 (April): 305-14. White, Gerald T. (1980). Billions for Defense: Government Financing by the Defense Plant Corporation During World War II (University, AL: University of Alabama Press). Worsley, Thomas Blanchard (194X) Wartime Economic Stabilization and the Efficiency of Government Procurement. Wright, Gavin (1974). “The Political Economy of New Deal Spending: An Econometric Analysis,” Review of Economics and Statistics, 56 (Feb.): 30-38. http://www.offutt.af.mil/geninfo/history/. 32 Data Appendix Spending Ltspc is the log of the county’s total spending (in thousands of dollars) to its 1940 population; Lfacpc is the log of facility investment (in thousands of dollars) per capita; Lspndpc is the log of major supply contracts (in thousands of dollars) per capita. Major supply contracts refers to “a prime contract involving a sum of $50,000 or more” awarded “by the War Department, the Navy Department, the Maritime Commission, the Treasury Procurement Division, and the foreign purchasing missions as reported to the War Production Board” over the June 1940 to September 1945 period. The values are net of cancellations and reductions. The contracts were assigned to “the location of the principal producing plants. The assignment of prime contracts to counties, however, may not be exact because of the difficulty involved in making proper assignments. Moreover, work may have been carried out in other counties through subcontracting.” They omit contracts that could not be assigned definitely as well as awards for foodstuffs and food processing. “Supply contracts for combat equipment include contracts for aircraft, ships, ordnance, and communication equipment.” “The value of facilities projects represent an estimated of the final cost of each project” undertaken by the WPB between June 1940 and June 1945 with a value of $25,000 or more. It includes facilities “financed by the Army, Navy, Maritime Commission, Defense Plant Corporation, Reconstruction Finance Corporation, and British Empire governments.” Industrial facilities “represent manufacturing facilities such as aircraft plants, shipbuilding yards, and metal working plants producing war material.” Military facilities “represent cantonments, airports, and other military installations for which direct outlays were made by the armed forces.” Source: County and City Data Book [United States] Consolidated File: County Data, 1944-1977 ICPSR. 7736; United States Department of Commerce. Bureau of the Census and 1947 County Databook, pp. xii- xiii. Population and Labor Force Variable Pop40 is the county’s population in April 1940; Pop40sq is Pop40 squared; Lpop40 is the log of Pop40. Prural measure the percent rural in 1940 and is calculated as (rural farm pop+rural non-farm pop)/pop40 Unemprt is the unemployment rate in 1940 and is calculated as 1-employment/labor force Lpoppc is the log of the ratio of the population is neighboring counties to the population in the subject county. If the neighboring county population is zero the log is set equally to the minimum value in the sample. The dummy variable Dnpoppc is set to one in this case; zero otherwise. Source: ICPSR. 7736 County and City Data Book [United States] Consolidated File: County Data, 1944- 1977 United States Department of Commerce. Bureau of the Census Geographic Variables Landarea is the number of square miles in the county. Coastal is an indicator for whether the county lies on coastline. Secure is an indicator for whether the county’s seat of government is 200 miles or more from US coastline of borders. Source: Sechrist, Robert P. Basic Geographic and Historic Data for Interfacing ICPSR Data Sets, 1620- 1983. [Computer file]. ICPSR version. Baton Rouge, LA: Robert P. Sechrist, LouisianaState University [producer], 1984. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2000. ICPSR 8159 . 33 Military Bases Army37 is an indicator for the presence of army post in the county in April 1937; Navy37 is the same for major navy shore establishments . NArmy37 and NNavy37 are indicator variables for the presence of such bases in neighboring counties. Sources: US War Department, Adjutant General’s Office, Army List and Directory April 20, 1937 (Washington DC: GPO, 1937) pp. 21-29; and US Navy Department, Bureau of Navigation, Navy Directory, April 1, 1937 (Washington DC: GPO, 1937) pp. 300-11. Industrial Capacity Air, Ship, Ord, and Auto are indicators variables for the presence, respectively, of aircraft, shipbuilding, ordnance (ammunition and firearms), and automobile (vehicle industry and motor vehicle body and parts industry) establishments in the county in the 1935 Census of Manufactures. NAir, NShip, NOrd, and Nauto are indicators for the presence of such establishments in neighboring counties. Source: Holleran, Owen Cobb. Industrial Market Data Handbook of the United States Bureau of Foreign and Domestic Commerce, Domestic commerce series no. 107; Washington,D.C. U.S. Dept. of Commerce, 1938 Manufacturing Wage-Earners Lmwepcm is the log of the manufacturing wage-earners in the county in 1939 Census of Manufactures to the 1940 population. If the ratio is zero the log is set equally to the minimum value in the sample. The dummy variable Dmwepc0 is set to one in this case; zero otherwise. Lnmwepc is the log of the manufacturing wage-earners in neighboring If the ratio is zero the log is set equally to the minimum value in the sample. The dummy variable DNmwepc is set to one in this case; zero otherwise. Source: ICPSR. 7736 County and City Data Book [United States] Consolidated File: County Data, 1944- 1977 United States Department of Commerce. The number of wage-earners in counties with 0-2 establishments is inferred by multiplying the average size (residual wage-earners/ residual establishments) in the state by the number of establishments in the county. Congressional Power Variables ssenpar is the senior senator’s party in 77th Congress (1942) (demo and fellow travelers=1); jsenpar is the same for junior senator lsten – the log of the combined tenure of senior and junior senators to 1942. scmt– indicator if a senator serves on Appropriations, Military, or Naval Affairs committee. slead – indicator (=1) if senator is President Pro Tem in 77th Congress. Glass of VA hparty – average party of house members in 77th Congress (demo and fellow travelers=1) lhten – the log of average tenure in house members in months to 1942 hcmt– indicator if a home member serves on Appropriations, Maritime, Military, or Naval Affairs committee. hlead – indicator if home member is Speaker (Rayburn), Majority Leader (McCormick), or Minority (Martin). Source: U.S. Congress, Official Congressional Directory Official Congressional Directory 77th Cong 2nd Sess, Corrected to Dec. 19. 1941, (Washington DC: U.S. G.P.O.) Election Variables Lecvpc is the log of the state’s number of Electoral College votes based on the 1940 realignment relative to the state population (in thousands) in 1940. Lecvpcsq is its square. 34 Lrat40 is the log of the number of house member to district’s 1940 population based on the 1940 realignment. Lrat40s is its square. Source: U.S. Congress, Official Congressional Directory Official Congressional Directory 78th Cong 1nd Sess, Corrected to May 14, 1943, (Washington DC: U.S. G.P.O.) Sdf50 is the absolute value of the difference of the state’s average democrat presidential vote share from 50 percent over the 1920-40 period. Dm24 is the county’s average democrat presidential vote share over the 1920-40 period; sddm24 is its standard deviation. sdDdm24 is the standard deviation of the state’s democrat presidential vote share over the 1920-40 period. Cdf50 is the absolute value of the difference of the congressional district’s average democrat presidential vote share from 50 percent over the 1932-40 period. cdm24 is the county’s average democrat congressional vote share over the 1920-40 period; sdcdm24 is its standard deviation. sdCDdm is the standard deviation of the congressional district’s democrat share over the 1932-40 period. Sources: Clubb, Jerome M., William H. Flanigan, and Nancy H. Zingale, Electoral Data for Counties in the United States: Presidential and Congressional Races, 1840-1972 [Computer file]. ICPSR ed. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [producer and distributor], 1986. ICPSR 8611 and Inter-university Consortium for Political and Social Research. United States Historical Election Returns, 1824-1968 [Computer file]. 2nd ICPSR ed. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [producer and distributor], 1999. ICPSR 1. We have corrected this data using state-level election reports. Note to be consistent with the spending data, the Virginia Independent Cities were assigned to the appropriate counties as follows: Charlottesville (Albemarle); Clifton Forge (Alleghany); Alexandria (Arlington); Staunton (Augusta); Lynchburg (Campbell); Petersburg, Hopewell (Dinwiddle);Wincester (Frederick); Richmond (Henrico); Martinsville (Henry); Williamsburg (James City); Radford (Montgomery); Suffolk (Nansemond); Danville (Pittsylvania); Roanoke City (Roanoke); Buena Vista (Rockbridge); Harrisonburg (Rockingham); Fredricksburg (Spotsylvania); Newport News (Warwick); Bristol (Washington); Norfolk City, Portsmouth, South Norfolk (Norfolk). 35 Table 1: Cumulative Military Procurement Spending by Quarter and Category, 1940-45 Total Supply Contract Facility Projects Expenditures Total Aircraft Ordnance Ships Comm. All Total Military Industrial Eqpmt Other (in million) (in million) (in million) (in million) (in million) (in million) (in million) (in million) (in million) (in million) 1940:II 1641 1641 1024 250 87 66 214 -- -- -- 1940:III 8007 6487 2288 1135 2423 93 548 1520 949 571 1940:IV 13774 10875 3823 2483 3167 172 1230 2899 1554 1345 1941:I 16867 12920 4094 3155 3713 225 1733 3947 2069 1878 1941:II 24478 19049 6801 4714 4549 393 2592 5429 2753 2676 1941:III 30477 22996 7724 6277 5379 423 3493 7481 3236 4245 1941:IV 39566 29945 10078 8511 6182 752 4722 9621 4205 5416 1942:I 67752 51777 17300 16327 9759 1175 7516 15975 6646 9329 1942:II 88184 67165 20696 21030 12738 2164 10837 21019 8598 12421 1942:III 106801 83533 26789 25316 15430 2681 13617 23268 9679 13589 1942:IV 120515 95810 29990 27834 16939 3374 17973 24705 10418 14287 1943:I 132093 105800 33072 29990 18004 4289 20745 26293 11273 15020 1943:II 151887 124743 40013 33323 21838 5463 24406 27144 11761 15383 1943:III 163640 135163 43346 35323 23375 5979 27440 28115 12136 15925 1943:IV 174443 145425 46168 37818 24640 6944 30155 28656 12309 16293 1944:I 186196 156708 51884 39843 25204 7628 32449 29126 12596 16476 1944:II 200610 170631 56472 43396 26660 8371 36032 29617 12872 16691 1944:III 208455 178158 57503 45232 27631 8717 39429 29935 13048 16833 1944:IV 216626 186083 59503 47652 28002 9128 42152 30181 13171 16956 1945:I -- -- -- -- -- -- -- -- -- -- 1945:II 227715 196004 59337 49468 29724 10188 47286 31771 13602 18110 Sources: Thomas B. Worsley Wartime Economic Stabilization and the Efficiency of Government Procurement Table 37, p. 394 Nelson A. Miller, State and Regional Market Indicators 1939-45, Econ. Series 60, p. 28-29. Table 2: Top Twenty Counties Ranked by Total and Per Capita Spending County and Total Spending County and Per Capita State (in thousands) State (in thousands) 1 Wayne, MI 11,753,727 Sarpy, NB 60.80 2 Los Angeles, CA 10,075,835 Orange, TX 38.01 3 Cook, IL 9,250,000 Sagadahoc, TX 26.83 4 Erie, NY 4,739,470 Mineral, NV 22.45 5 Cuyahoga, OH 4,666,873 Contra Costa, CA 20.75 6 Baltimore City, MD 3,914,239 Clark, IN 20.66 7 Hudson, NJ 3,722,262 Warwick, VA 18.86 8 Hartford, CT 3,677,341 Washtenaw, MI 17.82 9 Hamilton, OH 2,867,115 Schenectady, NY 16.21 10 Philadelphia, PA 2,817,073 Jackson, MS 15.85 11 Marion, IN 2,690,993 Clark, WA 14.12 12 Nassau, NY 2,616,102 St. Joseph, IN 12.20 13 Passaic, NJ 2,551,828 Columbia, PA 11.32 14 Oakland, MI 2,541,894 Douglas, KS 10.57 15 King, WA 2,489,474 Moore, TX 10.18 16 Kings, NY 2,388,732 Oakland, MI 10.00 17 New York, NY 2,319,014 New Hanover, NC 9.71 18 San Diego, CA 2,300,747 Sedgwick, KS 9.69 19 St. Louis City, MO 2,224,338 Kenosha, WI 9.59 20 Contra Costa, CA 2,084,314 St. Charles, MO 9.18 37 Table 3: Evolution of Procurement Policy and Agencies, 1939-45 1939 Spring Third revision of Industrial Mobilization Plan completed. 15 July Crowell Board on Educational Orders established. 9 August War Resources Board formed “to assist Army and Navy Munitions Board with plans for industrial mobilization.” 1 September Germany invades Poland. 24 November War Resources Board disbanded after issuing its report. 1940 16 May Roosevelt calls for 50,000 war planes. 28 May Roosevelt establishes National Defense Advisory Commission. 19 June Roosevelt forms War Cabinet by appointing Republicans Henry Stimson Secretary of War and Frank Knox Secretary of the Navy. 28 June Act to Expedite National Defense passes, allowing for negotiated contracts in place of competitive bidding. 22 August Reconstruction Finance Corporation forms Defense Plant Corporation. 29 December Roosevelt’s “Arsenal for Democracy” speech. 1941 7 January Office of Production Management established to replace NDAC. 1 March Senate creates “Truman Committee” to investigate defense program. 11 March Lend-Lease Act approved. 17 March OPM Plant Site Committee (later Board) established. 28 August Supply Priorities and Allocation Board formed with power over OPM. 3 December Production Requirements plan introduced. 7 December Pearl Harbor attacked; US enters War. 1942 16 January War Production Board formed to replace SPAB. 18 April War Manpower Commission established. 28 April Office of Price Administration “freezes prices.” 9 June Smaller War Plants Corporation established. 10 October WPB directs procurement agencies to avoid “Critical Labor Areas.” 38 2 November Controlled Materials Program announced. 1943 27 May Office of War Mobilization established to “harmonize government activities.” 5 November Truman Committee Report issued. 30 November WPB announces reconversion policy. 1944 3 October Office of War Mobilization and Reconversion established to replace OWM. 1945 8 May V-E Day. 2 September Formal V-J Day. 4 October WPB terminated, remaining functions transferred to Civilian Production Board. Sources: US Civilian Production Administration, Bureau of Demobilization, “Chronology of the War Production Board and Predecessor Agencies, August 1939 to November 1945,” Historical Reports on War Administration: War Production Board, Misc. Publ. No. 1 (June 20, 1945) and Industrial Mobilization for War: History of the War Production Board and Predecessor Agencies, 1940-45 Vol. I Program and Administration, Historical Reports on War Administration: War Production Board, General Study No. 1 (Washington, DC: GPO, 1947). 39 Table 4: National Political Balance Following 1932-48 Elections Presidential Elections Congress Electoral College Popular Vote Vote Percent House Seats Senate Seats Dem. Rep. Dem. Rep. Dem. Rep. Dem. Rep. Dem. Rep. Election 1932 472 59 22810 15759 57.4 39.7 310 117 60 35 1934 319 103 69 25 1936 523 8 27753 16675 60.8 36.5 331 89 76 16 1938 261 164 69 23 1940 449 82 27308 16675 54.7 33.4 268 162 66 28 1942 218 208 58 37 1944 432 99 25607 22015 53.4 45.9 242 190 56 38 1946 188 245 45 51 1948 303 189 24179 21991 49.6 45.1 263 171 54 42 Source: US Historical Statistics Y 79, 135, 204-209 40 Table 5: Summary Statistics of Variables Variable Symbol Obs Mean Std. Dev. Min Max Total Spending Ltspc 1787 -1.6672 2.0525 -8.1123 4.1077 Facility Investment Lfacpc 1064 -2.1569 1.9406 -8.1123 3.1114 Contract Spending Lspndpc 1569 -2.0835 2.0483 -6.8547 4.0708 Land Area Landarea 3071 968.12 1313.66 22 20131 Population 1940 Pop40 3071 42658.98 144048.80 42 4063342 squared Pop40sq 3071 2.260E+10 3.810E+11 1764 1.65E+13 log Lpop40 3071 9.8756 1.0611 3.7377 15.2175 Percent Rural 1940 Prural 3071 0.7706 0.2468 0 1 Unemployment Rate 1940 Unemprt 3071 13.6199 6.4207 0.6887 63.3913 Dummy=1 if Coastal Coastal 3071 0.1035 0.3047 0 1 Secure Secure 3071 0.6115 0.4875 0 1 Army base present Army37 3071 0.0485 0.2149 0 1 in neighbor Narmy37 3071 0.2042 0.4032 0 1 Navy base present Navy37 3071 0.0189 0.1361 0 1 in neighbor Nnavy37 3071 0.0866 0.2813 0 1 Aircraft est. present Air 3071 0.0147 0.1202 0 1 in neighbor Nair 3071 0.0664 0.2491 0 1 Shipbuilding est. present Ship 3071 0.0560 0.2300 0 1 in neighbor Nship 3071 0.1742 0.3794 0 1 Ordnance est. present Ord 3071 0.0081 0.0899 0 1 in neighbor Nord 3071 0.0446 0.2065 0 1 Automobile est. present Auto 3071 0.0765 0.2659 0 1 in neighbor Nauto 3071 0.2927 0.4551 0 1 Manufacturing wage-earners per capita Lmwepcm 3071 -4.6292 1.7499 -9.1371 -0.8129 dummy if zero Dmwepc0 3071 0.0368 0.1883 0 1 Wage-earners in neighboring counties LNmwepc 3071 -2.3568 2.6416 -10.1446 5.2862 dummy if zero DNmwepc 3071 0.0524 0.2229 0 1 Population in neighboring counties LNpoppc 3071 1.7389 1.5110 -2.8650 7.8195 dummy if zero DNpoppc 3071 0.0466 0.2107 0 1 Dummy =1 if Senior senator democrat Ssparty 3071 0.7457 0.4355 0 1 Junior senator democrat Jsparty 3071 0.6998 0.4584 0 1 Representative democrat Hparty 3071 0.6220 0.4838 0 1 Senate leadership Slead 3071 0.0326 0.1775 0 1 House leadership Hlead 3071 0.0033 0.0570 0 1 Senate committee Scmt 3071 0.6972 0.4596 0 1 House committee Hcmt 3071 0.2210 0.4130 0 1 Tenure of Senators Lsten 3071 4.4691 0.6125 2.4849 5.2883 Representatives Lhten 3071 4.0051 1.0223 -1.3863 5.9135 Electoral college votes per capita Lecvpc 3071 -5.4443 0.2596 -5.6587 -3.6019 squared Lecvpcsq 3071 29.7079 2.6201 12.9735 32.0213 Representatives per capita Lrrat40 3071 1.2520 0.2302 0.2824 2.2050 squared Lrrat40s 3071 1.6206 0.6007 0.0797 4.8622 41 Table 5: Summary Statistics of Variables continued Closeness of Election: State Presidential Sdf50 3071 0.1217 0.1005 0.0010 0.4585 Congressional Cdm50 3071 0.2145 0.1812 0.00050 0.5 Democrat Share: County Presidential Dm24 3071 0.5467 0.1916 0.096 1 County congressional Cdm24 3071 0.5865 0.2761 0 1 County St. Deviation: Presidential Sddm24 3071 0.1391 0.0612 0 0.3596 Congressional Sdcdm24 3071 0.0969 0.0616 0 0.3858 Election St. Deviation: Presidential SdSdm24 3071 0.1390 0.0463 0.0273 0.2389 Congressional SdCDdm 3071 0.0575 0.0465 0 0.3017 See Data Appendix for Definitions and Sources 42 TABLE 6: HECKMAN MAXIMUM LIKELIHIOOD ESTIMATES Presidential election variables Congressional election variables LTSPC LFACPC LCONTPC LTSPC LFACPC LCONTPC Log of Pop40 Coef. -0.2957 -0.6551 0.1546 -0.2855 -0.6887 0.1596 Std. Err. 0.0776 0.1043 0.0809 0.0748 0.1009 0.0785 Percent rural Coef. -2.9678 -1.2592 -2.4877 -2.9359 -1.2874 -2.5154 Std. Err. 0.2516 0.3582 0.2540 0.2523 0.3612 0.2545 Unemployment Rate Coef. -0.0091 0.0129 -0.0101 -0.0051 0.0215 -0.0093 Std. Err. 0.0075 0.0107 0.0075 0.0074 0.0107 0.0075 Coast Dummy Coef. 0.2810 0.1631 0.1926 0.2703 0.1348 0.1846 Std. Err. 0.1436 0.1783 0.1411 0.1449 0.1804 0.1430 Secure Coef. 0.0217 -0.0586 0.1259 0.0515 0.0541 0.1298 Std. Err. 0.1083 0.1522 0.1061 0.1060 0.1478 0.1042 Army base Coef. 0.2808 0.5251 0.0107 0.2596 0.5382 0.0043 Std. Err. 0.1705 0.1972 0.1651 0.1704 0.1973 0.1658 Neighbor Coef. 0.1133 0.0590 0.0040 0.1127 0.1172 -0.0083 Std. Err. 0.1160 0.1576 0.1116 0.1156 0.1567 0.1116 Navy Base Coef. 0.6878 0.9581 -0.2842 0.7607 0.9671 -0.2641 Std. Err. 0.2573 0.2824 0.2535 0.2572 0.2824 0.2551 Neighbor Coef. 0.3211 0.2498 0.3400 0.3353 0.3029 0.3288 Std. Err. 0.1675 0.2250 0.1576 0.1669 0.2233 0.1579 Aircraft Coef. 0.1956 0.2071 0.2392 0.1056 0.1190 0.1885 Std. Err. 0.2869 0.3184 0.2675 0.2888 0.3218 0.2707 Neighbor Coef. 0.1965 0.4912 0.0177 0.1715 0.4911 -0.0094 Std. Err. 0.1668 0.2220 0.1559 0.1675 0.2225 0.1573 Shipbuilding Coef. 0.4664 0.3367 0.5429 0.4458 0.3511 0.5207 Std. Err. 0.1726 0.2020 0.1648 0.1727 0.2025 0.1656 Ordnance Coef. 0.3244 0.2403 0.2335 0.4241 0.3498 0.2867 Std. Err. 0.3524 0.4090 0.3267 0.3535 0.4115 0.3296 Neighbor Coef. 0.4505 0.1680 0.3440 0.4609 0.1729 0.3712 Std. Err. 0.1427 0.1733 0.1330 0.1425 0.1728 0.1334 43 Automobile Coef. 0.1940 0.0956 0.1827 0.2285 0.1100 0.2362 Std. Err. 0.1124 0.1622 0.1081 0.1118 0.1616 0.1081 Wage-earners per capita Coef. 0.3336 -0.0452 0.6274 0.3417 -0.0311 0.6207 Std. Err. 0.0471 0.0633 0.0500 0.0471 0.0633 0.0501 Wage-earners in neighbors Coef. -0.3097 -0.1338 -0.1374 -0.3078 -0.1227 -0.1511 Std. Err. 0.0971 0.1305 0.0976 0.0959 0.1290 0.0975 Population in neighbors Coef. 0.1459 -0.0559 0.1652 0.1472 -0.0627 0.1803 Std. Err. 0.0595 0.0811 0.0618 0.0592 0.0807 0.0617 Senior senator democrat Coef. -0.3263 -0.1558 -0.2147 -0.3906 -0.2038 -0.2767 Std. Err. 0.1357 0.1811 0.1305 0.1381 0.1834 0.1319 Junior senator democrat Coef. 0.1971 0.2260 0.1904 0.2331 0.0323 0.3072 Std. Err. 0.1270 0.1713 0.1235 0.1239 0.1670 0.1210 Senate leadership Coef. -0.0893 -0.0018 0.0724 -0.1680 0.1273 -0.0171 Std. Err. 0.2331 0.3672 0.2237 0.2385 0.3746 0.2305 Senate tenure Coef. 0.0880 0.0243 0.0450 0.0933 0.1076 0.0212 Std. Err. 0.0762 0.1122 0.0736 0.0781 0.1139 0.0760 Senate committee Coef. -0.1328 -0.3054 -0.1371 -0.1223 -0.0682 -0.2053 Std. Err. 0.1151 0.1589 0.1134 0.1184 0.1639 0.1167 Representative democrat Coef. 0.0679 0.0906 0.1102 -0.0114 0.1014 0.0135 Std. Err. 0.1192 0.1642 0.1181 0.1202 0.1638 0.1187 House leadership Coef. 0.0924 -0.3712 0.0289 0.0779 -0.4151 0.1004 Std. Err. 0.6082 0.7553 0.5643 0.6067 0.7543 0.5651 House tenure Coef. -0.1210 -0.0308 -0.1427 -0.0787 -0.0429 -0.1057 Std. Err. 0.0425 0.0584 0.0425 0.0424 0.0579 0.0429 House committee Coef. 0.0140 -0.2201 0.1887 0.0079 -0.2870 0.1862 Std. Err. 0.0963 0.1330 0.0932 0.0966 0.1336 0.0945 Electoral college votes Coef. 10.2616 2.7461 12.2712 -2.4910 -4.3742 -1.3528 (Representatives) per capita Std. Err. 2.9565 3.6849 3.4483 1.1063 1.4561 1.1260 squared Coef. 0.9987 0.2232 1.2204 0.9637 1.6398 0.5472 Std. Err. 0.2928 0.3684 0.3380 0.4362 0.5742 0.4487 Closeness of Election Coef. -0.7330 -2.1446 0.0156 -0.5905 -0.0814 -0.1673 Std. Err. 0.6015 0.8892 0.5915 0.3709 0.5284 0.3774 44 Democrat vote share Coef. 0.7759 2.0464 -0.0932 1.1408 1.5896 0.1918 Std. Err. 0.4014 0.5993 0.4002 0.3109 0.4427 0.3183 County st. deviation Coef. 1.6140 3.2234 -0.4732 -0.2679 2.0450 -0.3755 Std. Err. 1.1263 1.5784 1.1430 0.8607 1.2294 0.8764 Election st. deviation Coef. -2.9998 -5.5681 1.2817 1.2887 -2.0823 0.9468 Std. Err. 1.5695 2.1937 1.5592 1.0351 1.3726 1.0348 SELECTION Land Area Coef. 6.45E-05 1.10E-04 2.16E-05 2.95E-05 9.86E-05 3.74E-06 Std. Err. 3.48E-05 3.26E-05 3.52E-05 3.21E-05 3.04E-05 3.23E-05 Population 1940 Coef. 2.90E-05 1.66E-05 3.42E-05 3.28E-05 1.76E-05 3.70E-05 Std. Err. 3.57E-06 2.17E-06 3.43E-06 3.50E-06 2.14E-06 3.37E-06 squared Coef. -1.11E-11 -4.17E-12 -1.27E-11 -1.25E-11 -4.43E-12 -1.38E-11 Std. Err. 5.20E-12 1.19E-12 3.80E-12 3.01E-12 1.10E-12 2.77E-12 Percent rural Coef. -2.2592 -1.7684 -2.1197 -2.1769 -1.7602 -2.0041 Std. Err. 0.2008 0.1728 0.1984 0.2003 0.1739 0.1992 Unemployment Rate Coef. -0.0030 -0.0076 0.0049 -0.0005 -0.0063 0.0075 Std. Err. 0.0050 0.0050 0.0052 0.0049 0.0049 0.0051 Coast Dummy Coef. 0.4083 0.5273 0.1046 0.4339 0.5525 0.1308 Std. Err. 0.1446 0.1185 0.1383 0.1457 0.1207 0.1397 Secure Coef. 0.1476 0.0647 0.1447 0.2033 0.1090 0.1831 Std. Err. 0.0859 0.0778 0.0885 0.0832 0.0757 0.0861 Army base Coef. 0.6933 0.5407 0.1298 0.6873 0.5307 0.1202 Std. Err. 0.3088 0.1926 0.2388 0.3056 0.1925 0.2386 Neighbor Coef. -0.0064 0.1537 -0.0708 -0.0264 0.1437 -0.0832 Std. Err. 0.0987 0.0878 0.1002 0.0979 0.0872 0.0993 Navy Base Coef. 1.4190 2.3616 -0.0060 1.3618 2.3736 -0.0036 Std. Err. 0.6200 0.6046 0.4231 0.6234 0.6008 0.4208 Neighbor Coef. -0.1362 0.0043 0.0118 -0.1293 0.0041 0.0090 Std. Err. 0.1473 0.1337 0.1465 0.1475 0.1341 0.1464 Aircraft Coef. 65.5919 0.3597 74.3440 74.1838 0.3822 80.5806 Std. Err. 3.39E+04 0.6757 3.37E+05 2.82E+04 0.6821 1.86E+05 45 Neighbor Coef. 0.3716 -0.1151 0.4077 0.4124 -0.0789 0.4363 Std. Err. 0.1923 0.1455 0.1928 0.1925 0.1460 0.1932 Shipbuilding Coef. 0.3254 0.3057 0.3856 0.3788 0.3031 0.4367 Std. Err. 0.2751 0.1758 0.2444 0.2736 0.1761 0.2431 Ordnance Coef. 5.5384 0.0791 6.3217 5.5002 0.0705 6.0648 Std. Err. 3.17E+04 0.4800 1.27E+06 2.75E+04 0.4790 4.67E+05 Neighbor Coef. 0.5046 0.2787 0.4663 0.4922 0.2823 0.4475 Std. Err. 0.2503 0.1784 0.2512 0.2480 0.1783 0.2476 Automobile Coef. 0.4914 0.2707 0.7889 0.4883 0.2794 0.7754 Std. Err. 0.3045 0.1477 0.3125 0.3018 0.1482 0.3094 Wage-earners per capita Coef. 0.3693 0.1263 0.4339 0.3678 0.1255 0.4357 Std. Err. 0.0284 0.0276 0.0305 0.0286 0.0276 0.0307 Wage-earners in neighbors Coef. -0.2081 -0.0633 -0.2169 -0.1696 -0.0372 -0.1918 Std. Err. 0.0771 0.0714 0.0794 0.0759 0.0711 0.0784 Population in neighbors Coef. 0.0780 -0.0176 0.1522 0.0854 -0.0218 0.1623 Std. Err. 0.0415 0.0397 0.0443 0.0413 0.0395 0.0441 Senior senator democrat Coef. -0.1183 -0.1286 -0.2184 -0.1443 -0.1294 -0.2402 Std. Err. 0.1272 0.1091 0.1297 0.1359 0.1145 0.1379 Junior senator democrat Coef. 0.0526 0.1256 0.1123 0.1194 0.1491 0.1490 Std. Err. 0.1079 0.0970 0.1105 0.1044 0.0942 0.1078 Senate leadership Coef. 0.0233 -0.1633 0.0894 0.1154 -0.1400 0.2176 Std. Err. 0.1786 0.1791 0.1760 0.1843 0.1841 0.1823 Senate tenure Coef. 0.1571 0.0568 0.1180 0.1893 0.0454 0.1659 Std. Err. 0.0610 0.0563 0.0622 0.0639 0.0584 0.0649 Senate committee Coef. 0.1773 0.0257 0.1228 0.1819 0.0312 0.1647 Std. Err. 0.0870 0.0823 0.0887 0.0915 0.0854 0.0939 Representative democrat Coef. 0.0750 -0.0511 0.1083 0.0458 -0.0792 0.0843 Std. Err. 0.0924 0.0863 0.0957 0.0942 0.0877 0.0979 House leadership Coef. 0.9122 0.3323 1.2733 0.7319 0.2246 1.1032 Std. Err. 0.8963 0.5803 1.0302 0.9105 0.5799 1.0285 House tenure Coef. -0.0042 -0.0195 0.0039 0.0161 -0.0076 0.0173 Std. Err. 0.0324 0.0306 0.0334 0.0322 0.0305 0.0332 46 House committee Coef. 0.0861 -0.0501 0.1518 0.0681 -0.0560 0.1515 Std. Err. 0.0796 0.0724 0.0810 0.0794 0.0728 0.0806 Electoral college votes Coef. 0.9438 0.0576 0.8972 -0.2572 1.3389 -1.2205 (Representatives) per capita Std. Err. 1.8931 1.8888 2.0795 0.8096 0.8394 0.8650 Squared Coef. 0.1193 0.0290 0.0850 0.1998 -0.4852 0.5455 Std. Err. 0.1900 0.1882 0.2076 0.3101 0.3226 0.3329 Closeness of Election Coef. -0.6819 -0.3725 -0.5385 -0.5925 0.0407 -0.9380 Std. Err. 0.4423 0.4221 0.4464 0.2815 0.2596 0.2874 Democrat vote share Coef. -0.1995 0.1616 -0.2650 0.2001 0.1390 0.3277 Std. Err. 0.2820 0.2774 0.2892 0.2295 0.2192 0.2370 County st. deviation Coef. -1.8947 -0.0383 -2.8436 -0.4183 -0.1378 -1.0049 Std. Err. 0.8078 0.7680 0.8414 0.6287 0.6081 0.6467 Election st. deviation Coef. -0.8592 -0.7656 -0.8297 -0.5703 0.2557 -0.8527 Std. Err. 1.1787 1.0997 1.2183 0.8638 0.7834 0.8713 Rho Coef. 0.0447 -0.1681 0.2840 0.0428 -0.1634 0.2732 Std. Err. 0.0678 0.0850 0.0778 0.0680 0.0871 0.0775 Sigma Coef. 1.6631 1.7818 1.5384 1.6624 1.7806 1.5450 Std. Err. 0.0279 0.0406 0.0296 0.0279 0.0405 0.0295 Lambda Coef. 0.0744 -0.2995 0.4369 0.0711 -0.2910 0.4220 Std. Err. 0.1128 0.1536 0.1230 0.1131 0.1571 0.1229 No. of Observations 3071 3071 3071 3071 3071 3071 Censored 1284 2007 1502 1284 2007 1502 Uncensored 1787 1064 1569 1787 1064 1569 Log Likelihood -4590.44 -3390.725 -3940.908 -4591.77 -3394.36 -3966.94 Notes: To save space, results for the presence of industries and military bases in neighboring counties are not displayed when statistically insignificantly different from zero. Constant and Dummies for zero logs also omitted. Note positive Ord and Air perfectly associated with positive spending. 47 TABLE 7 :TWO PART REGRESSION RESULTS Presidential election variables Congressional election variables LTSPC LFACPC LSPNDPC LTSPC LFACPC LSPNDPC Log of Pop40 Coef. -0.3040 -0.6100 0.0761 -0.2947 -0.6425 0.0812 Std. Err. 0.0794 0.0984 0.0809 0.0769 0.0956 0.0782 Percent rural Coef. -2.9208 -1.5275 -2.2344 -2.8931 -1.5464 -2.2873 Std. Err. 0.2400 0.3179 0.2518 0.2414 0.3194 0.2556 Unemployment Rate Coef. -0.0088 0.0109 -0.0096 -0.0049 0.0198 -0.0089 Std. Err. 0.0084 0.0120 0.0083 0.0084 0.0121 0.0084 Coast Dummy Coef. 0.2762 0.2033 0.1763 0.2656 0.1717 0.1708 Std. Err. 0.1438 0.1749 0.1465 0.1484 0.1807 0.1503 Secure Coef. 0.0195 -0.0646 0.1020 0.0491 0.0506 0.1111 Std. Err. 0.1081 0.1668 0.1050 0.1052 0.1612 0.1023 Army base Coef. 0.2801 0.5494 0.0219 0.2593 0.5599 0.0157 Std. Err. 0.1515 0.1535 0.1617 0.1533 0.1542 0.1621 Neighbor Coef. 0.1140 0.0833 0.0143 0.1138 0.1402 0.0036 Std. Err. 0.1130 0.1672 0.1104 0.1131 0.1660 0.1107 Navy Base Coef. 0.6864 1.0047 -0.2506 0.7593 1.0157 -0.2319 Std. Err. 0.2392 0.2286 0.2215 0.2471 0.2331 0.2221 Neighbor Coef. 0.3246 0.2569 0.3604 0.3390 0.3081 0.3507 Std. Err. 0.1666 0.2267 0.1613 0.1643 0.2234 0.1604 Aircraft Coef. 0.2151 0.1398 0.3689 0.1241 0.0554 0.3136 Std. Err. 0.2437 0.2553 0.2341 0.2408 0.2575 0.2367 Neighbor Coef. 0.1950 0.4827 0.0117 0.1693 0.4882 -0.0199 Std. Err. 0.1670 0.2160 0.1634 0.1656 0.2166 0.1631 Shipbuilding Coef. 0.4732 0.3247 0.5659 0.4521 0.3392 0.5419 Std. Err. 0.1582 0.1758 0.1670 0.1589 0.1793 0.1659 Ordnance Coef. 0.3352 0.1949 0.3060 0.4350 0.3062 0.3571 Std. Err. 0.2874 0.3813 0.2413 0.2969 0.3815 0.2435 Automobile Coef. 0.4601 0.1624 0.3854 0.4704 0.1664 0.4147 Std. Err. 0.1123 0.1742 0.1193 0.1119 0.1711 0.1202 48 Wage-earners per capita Coef. 0.3215 -0.0144 0.5487 0.3301 -0.0005 0.5454 Std. Err. 0.0531 0.0660 0.0574 0.0527 0.0652 0.0571 Wage-earners in neighbors Coef. 0.1459 -0.0609 0.1520 -0.3064 -0.1368 -0.1382 Std. Err. 0.0619 0.0814 0.0613 0.0953 0.1279 0.0896 Population in neighbors Coef. -0.3074 -0.1492 -0.1233 0.1469 -0.0663 0.1648 Std. Err. 0.0959 0.1292 0.0902 0.0605 0.0797 0.0603 Senior senator democrat Coef. -0.3213 -0.1927 -0.1817 -0.3871 -0.2354 -0.2534 Std. Err. 0.1295 0.1821 0.1231 0.1307 0.1792 0.1244 Junior senator democrat Coef. 0.1940 0.2563 0.1650 0.2290 0.0644 0.2778 Std. Err. 0.1281 0.1712 0.1218 0.1230 0.1649 0.1188 Senate leadership Coef. -0.0910 -0.0099 0.0589 -0.1713 0.1193 -0.0350 Std. Err. 0.2320 0.3823 0.2081 0.2362 0.3875 0.2126 Senate tenure Coef. 0.0865 0.0219 0.0406 0.0917 0.1013 0.0151 Std. Err. 0.0777 0.1245 0.0748 0.0808 0.1231 0.0789 Senate committee Coef. -0.1355 -0.3002 -0.1539 -0.1245 -0.0660 -0.2204 Std. Err. 0.1204 0.1681 0.1176 0.1222 0.1691 0.1190 Representative democrat Coef. 0.0663 0.0876 0.0968 -0.0113 0.0893 0.0135 Std. Err. 0.1196 0.1679 0.1126 0.1195 0.1633 0.1154 House leadership Coef. 0.0866 -0.3931 -0.0364 0.0754 -0.4405 0.0532 Std. Err. 0.3917 0.3176 0.5897 0.3697 0.3085 0.5821 House tenure Coef. -0.1209 -0.0370 -0.1438 -0.0791 -0.0469 -0.1086 Std. Err. 0.0432 0.0568 0.0435 0.0434 0.0550 0.0442 House committee Coef. 0.0122 -0.2102 0.1733 0.0064 -0.2760 0.1717 Std. Err. 0.0958 0.1369 0.0910 0.0977 0.1386 0.0940 Electoral college votes Coef. 10.2595 3.1527 12.6322 -2.4853 -4.3875 -1.2128 (Representatives) per capita Std. Err. 3.3947 3.1657 4.7497 1.1865 1.1983 1.1834 squared Coef. 0.9981 0.2643 1.2563 0.9596 1.6456 0.4836 Std. Err. 0.3334 0.3233 0.4582 0.4847 0.4944 0.4853 Closeness of Election Coef. -0.7220 -2.1640 0.0484 -0.5810 -0.0904 -0.0706 Std. Err. 0.6186 0.9036 0.6106 0.3955 0.5841 0.3850 49 Democrat vote share Coef. 0.7810 2.0146 -0.0417 1.1387 1.6057 0.1745 Std. Err. 0.4166 0.6513 0.4039 0.3157 0.4969 0.3126 County st. deviation Coef. 1.6932 3.1678 0.2683 -0.2416 2.0201 -0.1036 Std. Err. 1.1303 1.7036 1.0756 0.9184 1.2929 0.9117 Election st. deviation Coef. -3.0083 -5.7126 0.9829 1.2716 -1.9515 0.7382 Std. Err. 1.5844 2.3644 1.5370 1.0585 1.5192 1.0000 No. of Observations 1787 1064 1569 1787 1064 1569 R-squared = 0.343 0.1662 0.4477 0.344 0.1667 0.442 Root MSE = 1.681 1.8046 1.5411 1.6804 1.804 1.549 To save space, results for the presence of industries and military bases in neighboring counties are not displayed when statistically insignificantly different from zero. Constant and Dummies for zero logs also omitted. Standard Errors are Robust 50 Total Military Spending by County, 1940-1945 Total Spending (millions) > $200 (158) $75-$200 (135) $25-$75 (245) $0-$25 (1249) $0 (1286) Figure 1 51 Contract Spending by County, 1940-1945 Contract Spending (millions) > $200 (135) $75-$200 (103) $25-$75 (163) $0-$25 (1168) $0 (1504) Figure 2 52 Facility Spending by County, 1940-1945 Facility Spending (millions) > $100 (61) $50-$100 (86) $10-$50 (297) $0-$10 (620) $0 (2047) Figure 3 53 Per Capita Military Spending by County, 1940-1945 Spending Per Capita (thousands) > $5 (62) $1-$5 (347) $0-$1 (1378) $0 (1286) Figure 4 54