How Internal Constraints Shape Interest Group Activities TABLE 1. Differences-in-differences TABLE 2. Differences-in-Differences Estimates for Giveijt Estimates for Entryijt (1) (2) (1) (2) PAC_Rit x Indiv_Ri 0.0853** 0.0846* PAC_Rit x Indiv_Ri 0.0400** 0.0357* (0.00813) (0.00822) (0.00676) (0.00643) Donor-PAC fixed effect Y Y Donor-PAC fixed effect Y Y PAC-cycle fixed effect Y Y PAC-cycle fixed effect Y Y 2647085 3516248 1568054 2077662 Sample pure partisans all partisans Sample pure partisans all partisans Standard errors are clustered at the PAC level and reported in Standard errors are clustered at the PAC level and reported in parentheses. p<0.05,+p<0.01,**p<0.001 parentheses. p<0.05,#p<0.01,p<0.001 this indicator for donor entries,I expect the estimated opposite effect holds for pure partisan donors who sup- coefficient on Indiv_Rix PAC_Rit to be positive. port the opposite party. At the same time,active donors might be more likely One should note that limitations of campaign fi- to permanently stop donating following a decline in the nance records may considerably attenuate the esti- share of PAC contributions to copartisan candidates mates shown in Table 1.Recall that the outcome vari- To capture such exit effect,I construct Exitiit such that able,Giverjt,equals 0 whenever donor i did not make it equals 0 if donor i gave an itemized donation to PAC any itemized donation to PAC j during cycle t.This 4号元 j in cycle t;equals 1 if donor i permanently stopped could occur either because donor i chose not to do- making such donations to PAC j since cycle t;and is nate,or because donor i was simply ineligible to give coded as NA for all other cycles.Since exits should be (e.g.,donor i had not joined the parent organization of inversely related to donors'willingness to give,I expect PAC j in cycle t or had already left the organization by Indiv_Ri x PAC_Rit to have a negative estimated coef- then).Because the two scenarios are indistinguishable ficient when I substitute the outcome variable in Spec- in campaign finance records,the coefficient estimates ification 1 with this indicator for donor exits. for the interaction term PAC_Rit x Indiv_Ri reported in Table 1 may be biased towards zero by the inclusion Results of donor-cycles during which donors were ineligible to give and hence were not "treated"by changes in the 是 Consistent with my hypothesis,donors are less likely to allocations of PAC contributions across parties. give itemized donations to access-seeking PACs when There is suggestive evidence that the treatment ef- the share of PAC contributions to copartisan recipi- fects shown in Table 1 have risen over time.When I ents declines.Table 1 presents the coefficient estimates allow for a linear time trend in the treatment effect for two versions of Specification 1,where the outcome B in Specification 1,the relationship between donors' variable is Giverit,an indicator for whether donor i gave rates of itemized giving and the share of PAC con- any itemized donation to PACjduring cycle t.The first tributions to donors'copartisan recipients appears to column shows estimate treatment effect on the subset have strengthened over time (statistically significant of PAC donors who are "pure partisans,"i.e..they each at the 10%level).As shown in Section A.2.12 of the have only given to recipients of one party throughout online appendix,the estimated treatment effects on their donation histories.In comparison,the second col- donors'rates of giving have nearly doubled between umn use "all partisans,"i.e.,all donors that have given 1990 and 2016.This may have resulted from a concur- direct donations to recipients of either party. rent rise in polarization at both elite and mass levels In both cases,the estimated treatment effect B is sta- (McCarty,Poole,and Rosenthal 2016;Pew Research tistically significant and positive.Consider this base- Center 2017). line:pure partisan donors give itemized donations to The impact on donor behavior due to changes in access-seeking PACs at an average rate of 0.122 per access-seeking PACs'support for donors'preferred 四 cycle.Using column 1 of Table 1 as reference,the ab- parties is not only immediate,but also persistent.By solute rate of giving for pure partisan donors aligned substituting the outcome variable in Specification 1 with one party declines by 0.00839,or a 6.89%reduc- with Entryit,an indicator for donor entries,I can de- tion from the baseline.for a one-standard deviation de- tect whether donors who had not started giving(item- crease in the residualized share of PAC contributions ized donations)to access-seeking PACs would be more to the same party (19.7 percentage points).15 And the likely to do so when the share of PAC contributions to copartisan recipients rose.As Table 2 shows,this pre- diction is true regardless of whether I restrict to pure 15 Unless noted otherwise.I use to the standard deviation of partisan donors or not.Similarly,replacing the out- (PAC_R-PAC_Rj)across PACs j and cycles t to conduct coun- come variable in Specification 1 with Exiti,an indi- terfactual analyses,following Mummolo and Peterson(2018). cator for donor exits,allows me to examine whether 799How Internal Constraints Shape Interest Group Activities TABLE 1. Differences-in-differences Estimates for Giveijt (1) (2) PAC_Rjt × Indiv_Ri 0.0853∗∗∗ 0.0846∗∗∗ (0.00813) (0.00822) Donor-PAC fixed effect Y Y PAC-cycle fixed effect Y Y N 2647085 3516248 Sample pure partisans all partisans Standard errors are clustered at the PAC level and reported in parentheses. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 this indicator for donor entries, I expect the estimated coefficient on Indiv_Ri × PAC_Rjt to be positive. At the same time, active donors might be more likely to permanently stop donating following a decline in the share of PAC contributions to copartisan candidates. To capture such exit effect, I construct Exitijt such that it equals 0 if donor i gave an itemized donation to PAC j in cycle t; equals 1 if donor i permanently stopped making such donations to PAC j since cycle t; and is coded as NA for all other cycles. Since exits should be inversely related to donors’ willingness to give, I expect Indiv_Ri × PAC_Rjt to have a negative estimated coef￾ficient when I substitute the outcome variable in Spec￾ification 1 with this indicator for donor exits. Results Consistent with my hypothesis, donors are less likely to give itemized donations to access-seeking PACs when the share of PAC contributions to copartisan recipi￾ents declines.Table 1 presents the coefficient estimates for two versions of Specification 1, where the outcome variable is Giveijt, an indicator for whether donori gave any itemized donation to PAC j during cycle t. The first column shows estimate treatment effect on the subset of PAC donors who are “pure partisans,” i.e., they each have only given to recipients of one party throughout their donation histories. In comparison, the second col￾umn use “all partisans,” i.e., all donors that have given direct donations to recipients of either party. In both cases, the estimated treatment effect βˆ is sta￾tistically significant and positive. Consider this base￾line: pure partisan donors give itemized donations to access-seeking PACs at an average rate of 0.122 per cycle. Using column 1 of Table 1 as reference, the ab￾solute rate of giving for pure partisan donors aligned with one party declines by 0.00839, or a 6.89% reduc￾tion from the baseline, for a one-standard deviation de￾crease in the residualized share of PAC contributions to the same party (19.7 percentage points).15 And the 15 Unless noted otherwise, I use to the standard deviation of (PAC_Rjt − PAC_Rj) across PACs j and cycles t to conduct coun￾terfactual analyses, following Mummolo and Peterson (2018). TABLE 2. Differences-in-Differences Estimates for Entryijt (1) (2) PAC_Rjt × Indiv_Ri 0.0400∗∗∗ 0.0357∗∗∗ (0.00676) (0.00643) Donor-PAC fixed effect Y Y PAC-cycle fixed effect Y Y N 1568054 2077662 Sample pure partisans all partisans Standard errors are clustered at the PAC level and reported in parentheses. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 opposite effect holds for pure partisan donors who sup￾port the opposite party. One should note that limitations of campaign fi￾nance records may considerably attenuate the esti￾mates shown in Table 1. Recall that the outcome vari￾able, Giveijt, equals 0 whenever donor i did not make any itemized donation to PAC j during cycle t. This could occur either because donor i chose not to do￾nate, or because donor i was simply ineligible to give (e.g., donor i had not joined the parent organization of PAC j in cycle t or had already left the organization by then). Because the two scenarios are indistinguishable in campaign finance records, the coefficient estimates for the interaction term PAC_Rjt × Indiv_Ri reported in Table 1 may be biased towards zero by the inclusion of donor-cycles during which donors were ineligible to give and hence were not “treated” by changes in the allocations of PAC contributions across parties. There is suggestive evidence that the treatment ef￾fects shown in Table 1 have risen over time. When I allow for a linear time trend in the treatment effect β in Specification 1, the relationship between donors’ rates of itemized giving and the share of PAC con￾tributions to donors’ copartisan recipients appears to have strengthened over time (statistically significant at the 10% level). As shown in Section A.2.12 of the online appendix, the estimated treatment effects on donors’ rates of giving have nearly doubled between 1990 and 2016. This may have resulted from a concur￾rent rise in polarization at both elite and mass levels (McCarty, Poole, and Rosenthal 2016; Pew Research Center 2017). The impact on donor behavior due to changes in access-seeking PACs’ support for donors’ preferred parties is not only immediate, but also persistent. By substituting the outcome variable in Specification 1 with Entryijt, an indicator for donor entries, I can de￾tect whether donors who had not started giving (item￾ized donations) to access-seeking PACs would be more likely to do so when the share of PAC contributions to copartisan recipients rose. As Table 2 shows, this pre￾diction is true regardless of whether I restrict to pure partisan donors or not. Similarly, replacing the out￾come variable in Specification 1 with Exitijt, an indi￾cator for donor exits, allows me to examine whether 799 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