Appendix 3 _______________________________________

Statistical Results

TO EVALUATE THE INFLUENCE of the presidential campaign on voter decision making, we compare the effect of cross-pressures across levels of campaign exposure or in different campaign contexts. Because the dependent variable is a binary outcome (1 = voted for opposing party candidate, 0 = voted for own party candidate), we have estimated the effects in separate models. We do not estimate an interaction between cross-pressures and campaign exposure because it is not possible in nonlinear models (e.g., logit model) to evaluate an interaction effect simply by looking at the sign, magnitude, or statistical significance of the coefficient on the interaction term. For instance, see discussion in Ai and Norton, “Interaction Terms in Logit and Probit Models.” For easier interpretation of the results, we have estimated the substantive effects across all relevant models for the same respondent type (based on global means and modes) using Clarify software to calculate the 90 percent confidence bounds around those estimates. We also calculate the Wald chi-squared statistic to determine if the difference in estimated coefficients across the low-exposure and high-exposure groups is statistically significant (assuming residual variation is the same across groups).

TABLE A1 Model Results Used to Calculate Substantive Effects Reported in Figure 4.2
	Non-battle	Battle	No Visits	Many Visits	Less Attentive	Most Attentive	Less Aware	Most Aware	Non-battle (Most Attentive)	(Most Battle Attentive)
Constant	−2.64	−4.57	−3.52	−4.06	−2.24	−3.27	−2.13	−4.32	−3.5	−1.1
	(.58)	(1.08)	(1.13)	(.91)	(.91)	(.97)	(.84)	(.88)	(1.16)	(.1.73)
Strength of Partisanship	−1.78	−2.15	−1.85	−2.40	−1.61	−2.03	−1.25	−2.27	−2.3	−3.5
	(−.29)	(.52)	(0.52)	(0.57)	(0.54)	(0.48)	(0.45)	(0.40)	(.498)	(1.06)
Age	0.09	0.05	0.18	0.06	−0.08	0.03	−0.11	0.21	0.3	−0.5
	(0.08)	(0.11)	(.14)	(.10)	(.14)	(.12)	(.12)	(.09)	(.154)	(0.272)
Minority	0.09	−0.86	0.09	0.02	−0.30	−0.29	−0.30	0.42	0.2	−0.9
	(.29)	(.69)	(.67)	(.59)	(.51)	(.61)	(.44)	(.44)	(.539)	(0.961)
Female	0.12	−0.80	0.36	−1.08	0.20	−0.50	−0.10	0.14	−0.3	−2.2
	(.26)	(.42)	(.44)	(.39)	(.42)	(.48)	(.38)	(.33)	(.475)	(0.996)
Education	0.03	0.66	0.20	0.49	−0.08	0.30	0.19	0.22	0.1	0.6
	(.13)	(.20)	(.21)	(.18)	(.24)	(.20)	(.18)	(.18)	(.246)	(0.417)
Political Knowledge	−0.09	−0.07	−0.20	0.06	0.26	−0.35	0.06	−0.06	−0.4	−0.7
	(.12)	(.23)	(.18)	(.21)	(.22)	(.23)	(.19)	(.20)	(.246)	(0.423)
Cross-pressures Scale	3.44	5.49	3.68	4.96	2.19	7.21	1.56	5.61	6.4	12.3
	(.72)	(1.09)	(1.32)	(.98)	(1.32)	(1.29)	(1.20)	(.86)	(1.29)	(3.19)
Wald Chi-square Statistic	2.45		0.61		7.38		7.53		2.95
Cross-pressured Dummy	0.83	2.31	0.82	2.00	0.30	2.77	0.21	1.78
	(.25)	(.57)	(.44)	(.49)	(.44)	(.56)	(.36)	(.35)
Wald Chi-square Statistic	5.75		3.26		11.98		9.88
N	1325	570	516	600	381	758	444	1108	545	213
LR Chi-square	74.22	81.82	31.90	57.89	15.09	39.09	15.60	93.30	65.92	51.10
Pseudo R-square	0.14	0.29	0.16	0.28	0.08	0.31	0.03	0.28	0.31	0.50

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting a vote for the opposing party candidate. Sample limited to self-identified partisans only (leaners not included). Included in the table are the coefficients, standard errors, and Wald chi-square statistic for both the cross-pressures scale measure (percent of important cross-pressured issues) and the cross-pressured dummy variable (1 = greater than mean, the estimate used to calculate total effects on 2004 election outcome). Control variables and model fit statistics are reported for model with cross-pressures scale. Data source is the 2004 Blair Center Election Survey.

TABLE A2 Model Results Used to Calculate Substantive Effects Reported in Figure 4.3
	Non-battle	Battle	No Visits	Many Visits	Less Aware	Most Aware	Less Attentive	Most Attentive	Non-battle (Undecided)	Battle (Undecided)
Constant	−1.22	−2.45	−1.09	−1.88	−2.05	−2.90	−1.08	−2.46	−1.59	−0.28
	(.35)	(.55)	(.46)	(.40)	(.22)	(.50)	(.53)	(.67)	(.69)	(1.25)
Strength of Partisanship	−1.79	−1.74	−1.57	−1.80	−1.86	−1.08	−1.51	0.05	−1.22	−2.37
	(.17)	(.27)	(.21)	(.20)	(.17)	(.32)	(.32)	(.15)	(.41)	(.81)
Age	−0.02	0.05	0.02	0.00	0.04	−0.04	0.05	−0.10	0.05	−0.15
	(.04)	(.06)	(.06)	(.05)	(.04)	(.09)	(.07)	(.08)	(.08)	(.16)
Minority	−0.34	−0.57	−0.26	−0.34	−0.64	0.41	−0.54	0.09
	(.18)	(.45)	(.24)	(.23)	(.21)	(.37)	(.31)	(.32)
Female	0.05	−0.17	−0.04	−0.04	0.05	−0.15	0.01	0.33	0.23	−1.37
	(.14)	(.21)	(.18)	(.16)	(.14)	(.29)	(.22)	(.24)	(.27)	(.52)
Education	−0.30	−0.11	−0.32	−0.19			−0.37	−1.75	−0.11	−0.28
	(.08)	(.13)	(.11)	(.09)			(.13)	(.25)	(.16)	(.31)
Cross-pressures Scale	2.51	3.28	1.37	3.02	1.74	3.62	1.75	3.26	1.95	4.14
	(.28)	(.43)	(.28)	(.31)	(.23)	(.56)	(.40)	(.47)	(.55)	(1.06)
Wald Chi-square Statistic	2.27		15.77		9.73		6.02		3.35
N	2553	1176	1520	2016	2444	746	791	1280	344	145
LR Chi-square	292.6	137.19	116	246.78	244.56	76.72	66.38	127.44	27.22	39.19
Pseudo R-square	0.17	0.18	0.12	0.18	0.14	0.18	0.10	0.19	0.07	0.27

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting a vote for the opposing party candidates. Sample limited to self-identified partisans only (Independent leaners not included). Data source is the 2000 Knowledge Networks Election Study.

TABLE A3 Model Results Used to Calculate Substantive Effects Reported in Figures 4.5 and 4.6.
	Conventions		Debates
	Transition	Defection	Transition	Defection
Constant	0.21	−1.18	−0.69	−2.09
	(.27)	(.67)	(.35)	(.93)
Strength of Partisanship	−0.68	−1.03	−0.64	−0.85
	(.06)	(.31)	(.09)	(.42)
Age	−0.02	−0.19	−0.16	−0.19
	(.04)	(.09)	(.05)	(.12)
Minority	−0.12	−0.79	0.08	−0.05
	(.17)	(.45)	(.26)	(.65)
Female	0.20	0.07	0.12	0.40
	(.11)	(.27)	(.16)	(.42)
Education	−0.16	−0.40	−0.05	−0.28
	0.07	0.16	0.09	0.22
Previous Volatility			1.40	2.89
			(.27)	(.67)
Cross-pressured Dummy	0.83	1.16	0.99	0.87
	(.16)	(.27)	(.24)	(.41)
N	1918	1313	1663	777
LR Chi-square	172.36	55.07	104.80	32.37
Pseudo R-square	0.08	0.11	0.09	0.13

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting any change in vote choice (transition) or a change to support the opposing party candidate (defection) from before to after the event. Sample restricted to respondents interviewed before the event and with 10 days after the event; transition model includes partisans and Independents and defection model limited to partisans who did not already support the opposing party candidate. Previous volatility is an indicator if the individual had previously changed vote choice prior to pre-event interview. Data source is the 2000 Knowledge Networks Election Study.

TABLE A4 Model Results Used to Calculate Substantive Effects Reported in Figure 5.1
	All White Democrats		Southern White Democrats
	1960	1968	1960	1968
Constant	−2.13	−1.38	−0.99	−1.27
	(.64)	(.78)	(.86)	(.159)
Strength of Partisanship	−1.33	−1.40
	(.28)	(.32)
Age	0.02	0.01	0.00	−0.02
	(−.01)	(.01)	(.01)	(.02)
Female	0.33	0.04	0.23	0.70
	(.27)	(.30)	(.42)	(.69)
Education	0.08	0.13	0.05	0.24
	(.16)	(.18)	(.21)	(.33)
Cross-pressured, Racial Issue	0.66	2.70	0.42	2.27
	(.52)	(.83)	(.57)	(1.27)
N	390	321	117	58
LR Chi-square	31.20	39.90	1.3	7.3
Pseudo R-square	0.08	0.11	0.01	0.11

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting defection. Sample restricted to white Democrats (including Independent leaners). Cross-pressures measure is an indicator if the individual volunteered a racial policy dislike about the Democratic candidate or party or a racial policy like about the Republican candidate or party. Data source is the American National Election Study cumulative file.

TABLE A5 Model Results Used to Calculate Substantive Effects Reported in Figure 5.3
	1972	1976
Constant	−2.67	−3.24
	(1.01)	(1.15)
Strength of Partisanship	−0.84	−0.85
	(.31)	(.36)
Age	0.03	0.01
	(.01)	(.01)
Female	−0.24	0.63
	(.26)	(.32)
Education	0.00	0.11
	(.01)	(.08)
Political Knowledge	0.05	0.14
	(.16)	(.18)
Cross-pressured, Busing	1.80	0.32
	(.57)	(.55)
N	280	269
LR Chi-square	42.89	15.33
Pseudo R-square	0.11	0.05

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting defection. Sample restricted to white Democrats (including leaners) who had the same opinion on busing in both 1972 and 1976. Cross-pressures measure is an indicator of whether the individual was opposed to busing to achieve integration. Data source is the American National Election 1972–1976 panel.

TABLE A6 Model Results Used to Calculate Substantive Effects Reported in Figure 5.5
	1964	1968	1972	1976	1980	1984	1988	1992	1996	2000	2004
Constant	−4.74	−0.48	−0.99	−2.62	−1.97	−3.31	−2.38	−8.07	−6.68	0.54	−5.42
	(.771)	(.705)	(.623)	(.843)	(1.08)	(1.43)	(1.17)	(1.62)	(1.72)	(1.9)	(1.91)
Strength of Partisanship	−1.39	−1.35	−1.09	−0.75	−1.86	−1.08	−2.22	−2.08	−1.42	−1.33	−1.34
	(.319)	(.248)	(.236)	(.317)	(.408)	(.484)	(.584)	(.634)	(.659)	(.601)	(.635)
Age	0.01	0.00	0.02	−0.01	0.00	0.01	−0.02	0.00	0.01	−0.03	−0.01
	(.009)	(.007)	(.007)	(.008)	(.009)	(.013)	(.011)	(.013)	(.013)	(.016)	(.013)
Female	0.11	−0.21	−0.04	0.38	0.24	0.37	0.41	0.84	−0.22	0.20	0.99
	(.278)	(.22)	(.197)	(.258)	(.311)	(.402)	(.401)	(.484)	(.451)	(.495)	(.569)
Education	0.53	−0.09	−0.18	0.26	0.04	−0.11	−0.13	0.60	0.49	−0.50	0.08
	(.164)	(.14)	(.123)	(.163)	(.197)	(.269)	(.235)	(.284)	(.288)	(.344)	(.347)
Political Knowledge		−0.02	−0.15	0.09	−0.15	0.38	0.19	0.34	0.22	−0.22	0.82
		(.11)	(.11)	(.144)	(.19)	(.266)	(.211)	(.262)	(.252)	(.232)	(.259)
Cross-pressures Scale, Racial	0.03	0.02	0.03	0.02	0.02	0.02	0.02	0.02	0.02	0.01	0.00
	(.004)	(.003)	(.004)	(.005)	(.005)	(.008)	(.006)	(.008)	(.008)	(.008)	(.009)
Cross-pressures Scale, Moral					0.11	−0.23	0.02	0.05	0.02	0.02	0.03
					(.321)	(.474)	(.008)	(.011)	(.009)	(.01)	(.011)
N	541	433	515	394	262	184	256	374	323	172	199
LR Chi-square	49.17	52.08	74.93	20.25	38.82	20.61	30.06	50.62	20.94	12.98	37.57
Pseudo R-square	0.153	0.11	0.12	0.05	0.15	0.14	0.21	0.29	0.14	0.13	0.26

Note: Bolded coefficients are statistically significant at p<.05. Reported are logit results of model predicting defection. Sample restricted to white Democrats (including leaners). Political knowledge is the interviewer assessment of respondent knowledge. Data source is the American National Election Study cumulative file.