Commerce,Coalitions,and Factor Mobility September 2002 TABLE 1. Probit Estimations for Senate Votes on Trade Bills-Class Model Estimation Result Effect of Individual Variables (Dependent Variable=Vote for Protection) on Probability of Vote for Protection 1824601875-19131922-371945-621970-941824601875-19131922-371945621970-94 Value of farm -0.84 -0.82* -1.57*2.84* -1.26 -0.46 -0.62 -0.64 0.82 -0.31 production 0.50) (0.38) (0.53) (0.73) (0.77) (0.09) (0.07) (0.06) (0.10) (0.10) Employment in 9.32* 16.02* 9.38* 3.00 2.11 0.53 0.69 0.74 -0.32 0.42 manufacturing (2.21) (2.64) 3.27 (3.40) (4.12) (0.04) (0.03) (0.04 (0.07) (0.27) Profits in -6.04 -8.69* -2.89* -2.02 0.08 0.12 0.68 0.64 -0.38 0.34 manufacturing (2.10) (2.03) (1.43) (2.01) (1.59) (0.38) (0.04) (0.11) (0.08) (0.30) N 372 532 367 280 382 log-likelihood -225.25 -324.51-219.96-121.49-241.43 Pseudo-R2 .1246 .1189 .1288.1329.0270 Estimations include constant and dummy variables for individual bills(not shown).Standard errors in parentheses.'p<.05;"p<.01. PEffects estimated for change in each variable from minimum(0)to maximum(1)values for equations including only that variable and bill dummies using Clarify(King,Tomz,and Wittenberg 2000). should be positively associated with votes for protec- The value of farm production is negatively associ- tion in the first period and most of the second period ated with votes for protection,as anticipated,in all but and negatively thereafter. the fourth period.The votes taken in the immediate Table 1 reports two sets of results.On the left are the post-1945 years may be anomalous in this regard due estimated coefficients and pseudo-R2statistics from the to the new rural reliance on farm support programs probit estimations of the class model in each period, introduced in the 1930s.The estimated effects of farm- which can be compared (see Table 5 below)with the ing on votes (shown on the right)are smallest in the results from the alternative industry-group model.On first and last periods;the largest negative effects appear the right,to give some idea of the magnitude of the in the periods between 1875 and 1937.Manufacturing different effects,are the first differences in the prob- employment is positively associated with protectionist ability of voting for protection when each of the class votes,as expected,although the results are again less variables changes from its theoretical minimum to its clear between 1945 and 1962,the postwar boom period theoretical maximum value (from 0 to 1).Interpreting for all kinds of U.S.manufacturing exports.While the the estimated coefficients in the full model(on the left) class model anticipates that owners of capital favored is rather difficult here because employment and profits protection up until at least 1914,the coefficients for in manufacturing are so highly collinear across states the profits variable in the first three periods are nega- (they are correlated at about 0.7 in each period).Both tive.Since employment and profits are highly collinear, directly reflect the size of the manufacturing sector in however,this may simply indicate that highly capital- each state and the separate effects of the different class intensive producers were less supportive of protection variables are thus difficult to discern.35 An interesting than others.The effects of profits on votes,calculated part of the problem here is that when both employment with employment excluded from the estimation (on and profits are included in the one model,the estimated the right),are positive until 1937,and largest between coefficients will also measure the effects of variation 1875 and 1937,as are the effects of employment on in labor and capital intensities in manufacturing pro- votes.36 duction (using more labor with the same amount of Table 2 presents the results of estimations for the capital,and vice versa).As a partial corrective here same set of votes on trade legislation in the Senate,but I have simply calculated the first differences for each now using indicators of the importance of exporting variable (on the right)when other class variables are and import-competing industries in each state as the excluded from the model.The separate effects are less explanatory variables.In line with a simple industry- important,in the end,than the overall performance of group model,we anticipate that the importance of the class model in each period and how it compares with the industry-group model,so this is not a crucial issue. 36 I have tried variants of the basic class model for the recent peri- ods that include measures of the skill level of the workforce in each state assuming,in line with Midford (1993)and Scheve and Slaughter (1998,2000),that skilled workers,viewed as a separate class,oppose protection.Yet models that include measures of the proportion of the 35 For a discussion,see Gujarati 1995,327-35.The problem is not state's adult population with high school diplomas or higher levels of just inefficiency,though the standard errors for the estimates more education perform no better than the basic specification in Table 1. than double when all three variables are included in the model rather In none of the estimations are the coefficients on these variables than one alone.It is also a question of effective sample size:There significant,and often they take the wrong (positive)sign.Since such are hardly any observations,for instance,in which state employment data are unavailable for previous periods,I have reported only the in manufacturing is high while state profits in manufacturing are low simplest model here to provide straightforward comparisons over (or vice versa). time. 600Commerce, Coalitions, and Factor Mobility September 2002 TABLE 1. Probit Estimations for Senate Votes on Trade Bills-Class Model Estimation Result (Dependent Variable = Vote for Pr~tection)~ 1824-60 1875-1913 1922-37 1945-62 1970-94 Value of farm -0.84 -0.82' -1.57" 2.84** -1.26 production (0.50) (0.38) (0.53) (0.73) (0.77) Employment in 9.32" 16.02" 9.38** 3.00 2.1 1 manufacturing (2.21) (2.64) (3.27) (3.40) (4.12) Profits in -6.04** -8.69" -2.89' -2.02 0.08 manufacturing (2.1 0) (2.03) (1.43) (2.01) (1.59) N 372 532 367 280 382 log-likelihood -225.25 -324.51 -21 9.96 -121.49 -241.43 ~ieudo-~~,1246 ,1189 ,1288 ,1329 ,0270 Effect of Individual Variables on Probability of Vote for Protectionb 1824-60 1875-1913 1922-37 1945-62 1970-94 -0.46 -0.62 -0.64 0.82 -0.31 (0.09) (0.07) (0.06) (0.10) (0.10) 0.53 0.69 0.74 -0.32 0.42 (0.04) (0.03) (0.04) (0.07) (0.27) 0.12 0.68 0.64 -0.38 0.34 (0.38) (0.04) (0.1 1) (0.08) (0.30) aEstimations include constant and dummy variables for individual bills (not shown). Standard errors in parentheses. *pi .05; **pi.Ol. bEffects estimated for change in each variable from minimum (0) to maximum (1) values for equations including only that variable and bill dummies using Clarify (King, Tomz, and Wittenberg 2000). should be positively associated with votes for protec￾tion in the first period and most of the second period and negatively thereafter. Table 1reports two sets of results. On the left are the estimated coefficients and pseudo- R~statistics from the probit estimations of the class model in each period, which can be compared (see Table 5 below) with the results from the alternative industry-group model. On the right, to give some idea of the magnitude of the different effects, are the first differences in the prob￾ability of voting for protection when each of the class variables changes from its theoretical minimum to its theoretical maximum value (from 0 to 1). Interpreting the estimated coefficients in the full model (on the left) is rather difficult here because employment and profits in manufacturing are so highly collinear across states (they are correlated at about 0.7 in each period). Both directly reflect the size of the manufacturing sector in each state and the separate effects of the different class variables are thus difficult to di~cern.~%n interesting part of the problem here is that when both employment and profits are included in the one model, the estimated coefficients will also measure the effects of variation in labor and capital intensities in manufacturing pro￾duction (using more labor with the same amount of capital, and vice versa). As a partial corrective here I have simply calculated the first differences for each variable (on the right) when other class variables are excluded from the model. The separate effects are less important, in the end, than the overall performance of the class model in each period and how it compares with the industry-group model, so this is not a crucial issue. 35 For a discussion. see Gujarati 1995, 327-35. The problem is not just inefficiency, though the standard errors for the estimates more than double when all three variables are included in the model rather than one alone. It is also a question of effective sample size: There are hardly any observations, for instance, in which state employment in manufacturing is high while state profits in manufacturing are low (or vice versa). The value of farm production is negatively associ￾ated with votes for protection, as anticipated, in all but the fourth period. The votes taken in the immediate post-1945 years may be anomalous in this regard due to the new rural reliance on farm support programs introduced in the 1930s. The estimated effects of farm￾ing on votes (shown on the right) are smallest in the first and last periods; the largest negative effects appear in the periods between 1875 and 1937. Manufacturing employment is positively associated with protectionist votes, as expected, although the results are again less clear between 1945 and 1962, the postwar boom period for all kinds of U.S. manufacturing exports. While the class model anticipates that owners of capital favored protection up until at least 1914, the coefficients for the profits variable in the first three periods are nega￾tive. Since employment and profits are highly collinear, however, this may simply indicate that highly capital￾intensive producers were less supportive of protection than others. The effects of profits on votes, calculated with employment excluded from the estimation (on the right), are positive until 1937, and largest between 1875 and 1937. as are the effects of employment on votes.36 Table 2 presents the results of estimations for the same set of votes on trade legislation in the Senate, but now using indicators of the importance of exporting and import-competing industries in each state as the explanatory variables. In line with a simple industry￾group model, we anticipate that the importance of 36 I have tried variants of the basic class model for the recent peri￾ods that include measures of the skill level of the workforce in each state assuming, in line with Midford (1993) and Scheve and Slaughter (1998,2000), that skilled workers, viewed as a separate class, oppose protection. Yet models that include measures of the proportion of the state's adult population with high school diplomas or higher levels of education perform no better than the basic specification in Table 1. In none of the estimations are the coefficients on these variables significant, and often they take the wrong (positive) sign. Since such data are unavailable for previous periods, I have reported only the simplest model here to provide straightforward comparisons over time