How Internal Constraints Shape Interest Group Activities level,are not sensitive to factors such as electoral Responsive Politics 2018).Section A.2.2 of the ap- competition or presidential cycles that would typically pendix contains descriptive summaries of the data affect electorally motivated donations.This suggests donors might not be taking their partisan or ideological preferences into account when giving to access-seeking Empirical Strategy PACs,although Lowry(2013)recognizes the limits to I employ the following difference-in-differences de- causal inference from ecological analyses. Sign:1 Min and You (2015)find that activist sharehold- ers (e.g.,public pension fund and union investors) target corporations whose political activities conflict Giveiit aij +tit +B(Indiv_Ri x PAC_Rit)+Eiit, with the partisan or ideological preferences of these 1) shareholders (e.g.,corporations that predominantly where i denotes an individual donor,j represents a donate to Republican politicians).This lends support PAC,and t indicates a 2-year federal election cycle.The to the conjecture that shareholders and other eligi- outcome variable,Givert,is an indicator that equals 1 if ble PAC donors might also constrain the fundraising and only if donor i gave any itemized donation to PACj of access-seeking PACs due to partisan or ideological during cycle t.2 On the right-hand side of Equation(1) disagreements. are two fixed effects,one for each donor-PAC dyad ( Bonica(2016a)shows that corporate executives,un- and another at the PAC-cycle level (it).In addition,the like the PACs they preside over,behave like ideologues main independent variable of interest is the product rather than investors when donating to candidates on of two variables:Indiv_Ri E[-1/2,1/2],the net share their own.Moreover.there is a remarkable degree of of donor i's lifetime donations to Republican (as op- heterogeneity in political ideologies even on the same posed to Democratic)candidates and party commit- corporate board.The prevalence of bipartisan board- tees,and PAC_Rit E[-1/2,1/2],the net share of PAC rooms,Bonica(2016a)argues,could account for why i's contributions to Republican(as opposed to Demo- 4号元 corporate PACs appear moderate in their contribution cratic)candidates and party committees during cycle patterns,for otherwise a clear partisan or ideological t.Indiv R;is a reliable measure of individual donors bias in PAC giving could antagonize elite donors.That partisan leanings because,as the left panel of Figure 2 being said,we do not know this for sure without a sys- shows,77%of all donors in my sample have given to tematic examination of how donor behavior depends only one party in their entire donation histories (so on the degree to which access-seeking PACs support Indiv R;mostly clusters at either-1/2 or 1/2).Barber. politicians that share the donors'partisan preferences Canes-Wrone,and Thrower(2017)also independently validate my method of classifying donors'partisanship. STUDY 1:DIFFERENCES-IN-DIFFERENCES As usual,I do not preclude the possibility that my mea- ANALYSIS OF CAMPAIGN FINANCE DATA sure of partisanship is a proxy for ideology.Note that I do not separately control for Indiv_Ri or PAC_Rit Using difference-in-differences analysis of campaign fi- because the donor-PAC fixed effect (a)absorbs the nance records,I first verify that patterns of real giving former while the PAC-cycle fixed effect (it)absorbs are consistent with my main hypothesis:access-seeking the latter.I cluster all standard errors at the PAC level 5.501g PACs lose donors with a partisan preference when giv- Since Indiv R;represents donor i's time-invariant ing to politicians of the opposite party.In particular, relative preference for the Republican Party,and such impact on donor behavior appears to have grown PAC_Rit measures PACi's relative support for the Re- in time,coinciding with a rise in polarization(Mc- publican Party in cycle t,their product captures the de- Carty,Poole,and Rosenthal 2016;Pew Research Center gree of alignment between PAC is contributions dur- 2017).Moreover,these effects also appear to be long- ing cycle t and donor i's partisan preference.Therefore, lasting:eligible but inactive donors are more likely to my hypothesis is that B>0 in Specification 1. start giving when the share of PAC contributions to I identify B by comparing within-person changes in their preferred party increases,and active donors are rates of giving by donors of different partisan leanings more likely to permanently stop giving when changes belonging to the same PAC.This becomes clear once in PAC contributions benefit the party they oppose. the donor-PAC fixed effect i and the PAC-cycle fixed I examine PACs that are coded as business PACs by effect rjt are differenced out of Specification 1,resulting OpenSecrets(Center for Responsive Politics 2018).101 in eys then construct a panel of individual-to-PAC as well as PAC-to-candidate/party committee contributions for cycles 1990-2016 using OpenSecrets data(Center for △，Givein-△Giveri=B(Indiv_.R:-Indiv_Rr) ×△PAC_Rit+㎡t, (2) 10 OpenSecrets categorizes PACs as business,labor,ideological,or others.I exclude a small minority of these business PACs that are not sponsored by a corporation,trade organization (such as the Mort- gage Bankers Association),or membership organization (such as the 11 I estimate this model using Stata package reghdfe developed by National Association of Realtors)to focus on access-seeking PACs as /:sony empirically defined in the existing literature(see,for example,Rom Section2 of the onine appendix.I discuss why one cao and Snyder 1994:Cox and Magar 1999:Bonica 2013:Drutman 2015) causally estimate treatment effects on amounts of giving due to post- See Section A.2.1 of the online appendix for details. treatment censoring. 797
How Internal Constraints Shape Interest Group Activities level, are not sensitive to factors such as electoral competition or presidential cycles that would typically affect electorally motivated donations. This suggests donors might not be taking their partisan or ideological preferences into account when giving to access-seeking PACs, although Lowry (2013) recognizes the limits to causal inference from ecological analyses. Min and You (2015) find that activist shareholders (e.g., public pension fund and union investors) target corporations whose political activities conflict with the partisan or ideological preferences of these shareholders (e.g., corporations that predominantly donate to Republican politicians). This lends support to the conjecture that shareholders and other eligible PAC donors might also constrain the fundraising of access-seeking PACs due to partisan or ideological disagreements. Bonica (2016a) shows that corporate executives, unlike the PACs they preside over, behave like ideologues rather than investors when donating to candidates on their own. Moreover, there is a remarkable degree of heterogeneity in political ideologies even on the same corporate board. The prevalence of bipartisan boardrooms, Bonica (2016a) argues, could account for why corporate PACs appear moderate in their contribution patterns, for otherwise a clear partisan or ideological bias in PAC giving could antagonize elite donors. That being said, we do not know this for sure without a systematic examination of how donor behavior depends on the degree to which access-seeking PACs support politicians that share the donors’ partisan preferences. STUDY 1: DIFFERENCES-IN-DIFFERENCES ANALYSIS OF CAMPAIGN FINANCE DATA Using difference-in-differences analysis of campaign finance records, I first verify that patterns of real giving are consistent with my main hypothesis: access-seeking PACs lose donors with a partisan preference when giving to politicians of the opposite party. In particular, such impact on donor behavior appears to have grown in time, coinciding with a rise in polarization (McCarty,Poole, and Rosenthal 2016;Pew Research Center 2017). Moreover, these effects also appear to be longlasting: eligible but inactive donors are more likely to start giving when the share of PAC contributions to their preferred party increases, and active donors are more likely to permanently stop giving when changes in PAC contributions benefit the party they oppose. I examine PACs that are coded as business PACs by OpenSecrets (Center for Responsive Politics 2018).10 I then construct a panel of individual-to-PAC as well as PAC-to-candidate/party committee contributions for cycles 1990–2016 using OpenSecrets data (Center for 10 OpenSecrets categorizes PACs as business, labor, ideological, or others. I exclude a small minority of these business PACs that are not sponsored by a corporation, trade organization (such as the Mortgage Bankers Association), or membership organization (such as the National Association of Realtors) to focus on access-seeking PACs as empirically defined in the existing literature (see, for example,Romer and Snyder 1994; Cox and Magar 1999; Bonica 2013; Drutman 2015). See Section A.2.1 of the online appendix for details. Responsive Politics 2018). Section A.2.2 of the appendix contains descriptive summaries of the data. Empirical Strategy I employ the following difference-in-differences design:11 Givei jt = αi j + τjt + β(Indiv_Ri × PAC_Rjt) + i jt, (1) where i denotes an individual donor, j represents a PAC, and t indicates a 2-year federal election cycle. The outcome variable, Giveijt, is an indicator that equals 1 if and only if donori gave any itemized donation to PAC j during cycle t. 12 On the right-hand side of Equation (1) are two fixed effects, one for each donor-PAC dyad (αij) and another at the PAC-cycle level (τ jt). In addition, the main independent variable of interest is the product of two variables: Indiv_Ri ∈ [−1/2, 1/2], the net share of donor i’s lifetime donations to Republican (as opposed to Democratic) candidates and party committees, and PAC_Rjt ∈ [−1/2, 1/2], the net share of PAC j’s contributions to Republican (as opposed to Democratic) candidates and party committees during cycle t. Indiv_Ri is a reliable measure of individual donors’ partisan leanings because, as the left panel of Figure 2 shows, 77% of all donors in my sample have given to only one party in their entire donation histories (so Indiv_Ri mostly clusters at either −1/2 or 1/2). Barber, Canes-Wrone, and Thrower (2017) also independently validate my method of classifying donors’ partisanship. As usual, I do not preclude the possibility that my measure of partisanship is a proxy for ideology. Note that I do not separately control for Indiv_Ri or PAC_Rjt because the donor-PAC fixed effect (αij) absorbs the former while the PAC-cycle fixed effect (τ jt) absorbs the latter. I cluster all standard errors at the PAC level. Since Indiv_Ri represents donor i’s time-invariant relative preference for the Republican Party, and PAC_Rjt measures PAC j’s relative support for the Republican Party in cycle t, their product captures the degree of alignment between PAC j’s contributions during cycle t and donor i’s partisan preference. Therefore, my hypothesis is that β > 0 in Specification 1. I identify β by comparing within-person changes in rates of giving by donors of different partisan leanings belonging to the same PAC. This becomes clear once the donor-PAC fixed effect αij and the PAC-cycle fixed effect τ jt are differenced out of Specification 1, resulting in tGivei jt − tGivei jt = β(Indiv_Ri − Indiv_Ri ) ×tPAC_Rjt + ξii jt, (2) 11 I estimate this model using Stata package reghdfe developed by Correia (2017). 12 In Section 2.11 of the online appendix, I discuss why one cannot causally estimate treatment effects on amounts of giving due to posttreatment censoring. 797 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000382
Zhao Li as PAC-specific time trends14 that could confound the FIGURE 3.Visual Representation of the estimate of B.Therefore,causal identification of the Difference-in-Differences Design treatment effect B in Specification 1 relies on a parallel- trends assumption:that within a given PAC over time observed rates of itemized giving across donors of dif- E[Giveit] E[Giveit R] ferent partisan leanings would have the same counter- factual trends if the allocation of PAC contributions across parties in each election cycle were held constant In Section A.2.5 of the online appendix,I show that estimates obtained from a modified Granger causality test are indeed consistent with the null hypothesis of no differential trends.Moreover,in the online appendix,I E[Giveijt D] show that estimates obtained from Specification 1 are robust to(1)regressing on lagged rather than same- ………………… cycle PAC contribution patterns (Section A.2.6),(2) using a plausible instrument for PAC contribution pat- terns (Section A.2.7),and (3)controlling additionally t t+1 Cycle for a proxy of potential confounding trends(Section PAC Rjt A.2.8).In all three cases,I obtain estimates that are qualitative identical and quantitatively similar to those from Specification 1. 1/2 小。：。。te。。n Because OpenSecrets'data spans 14 election cycles (Center for Responsive Politics 2018),I also explore two types of dynamic patterns.First,I examine whether 4号 the treatment effect,B.has changed over time by al- lowing it to have a linear time trend.Second.I test for t+1 Cycle persistence in the treatment effect on donor behavior. One such persistent effect is entry:eligible but inactive donors might be more likely to start donating to access- seeking PACs when a greater share of PAC contribu- tions support their preferred party.I construct Entryr which equals 1 if donor i's first itemized donation to where donor i and i'both belong to PAC j,and i= PACjoccurred in cycle t;equals 0 for all cycles prior to Areiit-Arer it.For a visual representation of the identi- t;and is coded as NA for all cycles after t.When sub- 是 fication strategy for B in Equation(2),consider a hypo- stituting the outcome variable in Specification 1 with thetical PAC j where,for simplicity,I assume that every PAC donor is either a pure Republican (i.e.,Indiv_Ri= 1/2)or a pure Democrat (i.e.,Indiv_R:=-1/2).Sup- more Republican leaning than the employees,(2)the executives can steer the company's PAC towards giving to Republican candidates, pose PAC j gave only to Democratic recipients dur- and(3)the likelihood of giving to the company's PAC depends solely ing cycle t (i.e.,PAC_Rit=-1/2)and exclusively to on one's income.Under these conditions,the executives will be more Republican recipients in the next cycle (PAC_Rit+1= likely to give to the PAC.At the same time,this PAC's contributions 1/2),as shown in the bottom half of Figure 3.If the will also appear more aligned with the executives'partisan prefer- average rate of Republican donors giving to PAC j ences.The estimate of B would be positive even though partisan changed by y between the two cycles,and the same av- preference,by assumption,plays no role in the decision to give to the company's PAC. erage rate for Democratic donors changed by x,then B 14 Not allowing time trends to vary by PAC could also cause spuri- =y-x is the causal effect of the degree of alignment ous correlations.For example,suppose donors'willingness to give to between PAC contributions and donors'partisan pref- access-seeking PACs depends only on the rates of investment returns to PAC contributions.Also,assume such returns are higher for the erences on rates of itemized donations,as shown in the energy sector when Republicans dominate the House of Represen- top half of Figure 3.Suppose B=0.10,the interpreta- tatives,and that the opposite holds for the technology sector.More- tion would be that as a result of a full-unit increase in over,suppose workers in the energy sector lean Republican while the share of PACcontributions to one party across con- technology workers lean Democratic.Then,if majority control over eys secutive cycles,the change in the absolute probability the House changes from Democratic to Republican across consecu- that donors who prefer said party will donate to the tive cycles t and t+1,two things will happen.First,because access- seeking PACs favor incumbents,PACs representing both sectors will PAC will be 10 percentage points higher than that for give a greater share of their contributions to Republicans in t+1 donors who prefer the opposite party. than int.Second,returns to PAC contributions will rise for the energy The two fixed effects in Specification 1,ai and tit, sector and fall for the technology sector.These factors together could cause donors in the energy sector to give at a higher rate to their control for all time-invariant donor attributes as well PACs in the second cycle than in the first,and the opposite could be true for donors in the technology sector.Without accounting for PAC-specific trends here,comparing within-person changes in rates 13 Time-invariant donor attributes could generate spurious correla. of giving would lead to the false conclusion that donors'willingness tions between the degree to which PACcontributions appear aligned to give depends on the degree of alignment between PAC activities with donors'partisan preferences and donors'rates of itemized giv- and donors'partisan preference (when there is no such effect by as- ing.Consider a hypothetical company where (1)the executives are sumption). 798
Zhao Li FIGURE 3. Visual Representation of the Difference-in-Differences Design where donor i and i both belong to PAC j, and ξii jt = ti jt − ti jt . For a visual representation of the identification strategy for β in Equation (2), consider a hypothetical PAC j where, for simplicity, I assume that every PAC donor is either a pure Republican (i.e.,Indiv_Ri = 1/2) or a pure Democrat (i.e.,Indiv_Ri = −1/2). Suppose PAC j gave only to Democratic recipients during cycle t (i.e., PAC_Rjt = −1/2) and exclusively to Republican recipients in the next cycle (PAC_Rj,t+1 = 1/2), as shown in the bottom half of Figure 3. If the average rate of Republican donors giving to PAC j changed by y between the two cycles, and the same average rate for Democratic donors changed by x, then β = y − x is the causal effect of the degree of alignment between PAC contributions and donors’ partisan preferences on rates of itemized donations, as shown in the top half of Figure 3. Suppose β = 0.10, the interpretation would be that as a result of a full-unit increase in the share of PAC contributions to one party across consecutive cycles, the change in the absolute probability that donors who prefer said party will donate to the PAC will be 10 percentage points higher than that for donors who prefer the opposite party. The two fixed effects in Specification 1, αij and τ jt, control for all time-invariant donor attributes13 as well 13 Time-invariant donor attributes could generate spurious correlations between the degree to which PAC contributions appear aligned with donors’ partisan preferences and donors’ rates of itemized giving. Consider a hypothetical company where (1) the executives are as PAC-specific time trends14 that could confound the estimate of β. Therefore, causal identification of the treatment effect β in Specification 1 relies on a paralleltrends assumption: that within a given PAC over time, observed rates of itemized giving across donors of different partisan leanings would have the same counterfactual trends if the allocation of PAC contributions across parties in each election cycle were held constant. In Section A.2.5 of the online appendix, I show that estimates obtained from a modified Granger causality test are indeed consistent with the null hypothesis of no differential trends. Moreover, in the online appendix, I show that estimates obtained from Specification 1 are robust to (1) regressing on lagged rather than samecycle PAC contribution patterns (Section A.2.6), (2) using a plausible instrument for PAC contribution patterns (Section A.2.7), and (3) controlling additionally for a proxy of potential confounding trends (Section A.2.8). In all three cases, I obtain estimates that are qualitative identical and quantitatively similar to those from Specification 1. Because OpenSecrets’ data spans 14 election cycles (Center for Responsive Politics 2018), I also explore two types of dynamic patterns. First, I examine whether the treatment effect, β, has changed over time by allowing it to have a linear time trend. Second, I test for persistence in the treatment effect on donor behavior. One such persistent effect is entry: eligible but inactive donors might be more likely to start donating to accessseeking PACs when a greater share of PAC contributions support their preferred party. I construct Entryijt which equals 1 if donor i’s first itemized donation to PAC j occurred in cycle t; equals 0 for all cycles prior to t; and is coded as NA for all cycles after t. When substituting the outcome variable in Specification 1 with more Republican leaning than the employees, (2) the executives can steer the company’s PAC towards giving to Republican candidates, and (3) the likelihood of giving to the company’s PAC depends solely on one’s income. Under these conditions, the executives will be more likely to give to the PAC. At the same time, this PAC’s contributions will also appear more aligned with the executives’ partisan preferences. The estimate of β would be positive even though partisan preference, by assumption, plays no role in the decision to give to the company’s PAC. 14 Not allowing time trends to vary by PAC could also cause spurious correlations. For example, suppose donors’ willingness to give to access-seeking PACs depends only on the rates of investment returns to PAC contributions. Also, assume such returns are higher for the energy sector when Republicans dominate the House of Representatives, and that the opposite holds for the technology sector. Moreover, suppose workers in the energy sector lean Republican while technology workers lean Democratic. Then, if majority control over the House changes from Democratic to Republican across consecutive cycles t and t + 1, two things will happen. First, because accessseeking PACs favor incumbents, PACs representing both sectors will give a greater share of their contributions to Republicans in t + 1 than in t. Second, returns to PAC contributions will rise for the energy sector and fall for the technology sector.These factors together could cause donors in the energy sector to give at a higher rate to their PACs in the second cycle than in the first, and the opposite could be true for donors in the technology sector. Without accounting for PAC-specific trends here, comparing within-person changes in rates of giving would lead to the false conclusion that donors’ willingness to give depends on the degree of alignment between PAC activities and donors’ partisan preference (when there is no such effect by assumption). 798 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000382