The paper
Sean J. Westwood (2015). The Role of Persuasion in Deliberative Opinion Change. Political Communication 32(4), 509–528.
doi:10.1080/10584609.2015.1017628 (opens in a new tab) · No public replication archive
Main text
p. 509, introduction
Published
Gastil, Black, & Moscovitz, 2008; Karpowitz, Mendelberg, & Shaker, 2012; Luc, 2013; Mendelberg, Karpowitz, & Goedert, 2013).
Correction
Gastil, Black, & Moscovitz, 2008; Karpowitz, Mendelberg, & Shaker, 2012; Luc, 2013; Mendelberg, Karpowitz, & Goedert, 2014).
Basis Mendelberg, Karpowitz and Goedert appeared in the American Journal of Political Science 58(2), 2014. The same correction applies on pp. 511 and 515 and in the reference list.
p. 510, introduction
Published
(see, for example, Kim, & Wyatt, Katz, 1999; Price, Cappella, & Nir, 2002)
Correction
(see, for example, Kim, Wyatt, & Katz, 1999; Price, Cappella, & Nir, 2002)
p. 510, introduction
Published
Results show that persuasive argumentation is both the largest and most consistent indicator of opinion change.
Correction
Results show that persuasive argumentation is the most consistent indicator of opinion change and has larger effects than knowledge change.
Basis In Table 3, distance from the initial group attitude has a larger coefficient than interactions from more liberal peers on free trade (−0.30 against 0.21) and protectionism (−0.37 against 0.32). See the correction to p. 520.
p. 511, Studying Deliberative Discussion and Persuasion and Opinion Change
Published
(Price, Nir, & Cappella, 2005, p. 5) [...] (Karpowitz et al., 2012; Mendelberg et al., 2013; Price et al., 2005) [...] (Allport, l935; Kelman, 1961) [...] (Festinger, 1964; Howland, Kelley, & Janis, 1954; McGuire, 1968)
Correction
(Price, 2009, p. 41) [...] (Karpowitz et al., 2012; Mendelberg et al., 2014; Price et al., 2005) [...] (Allport, 1935; Kelman, 1961) [...] (Festinger, 1964; Hovland, Janis, & Kelley, 1953; McGuire, 1968)
Basis The quotation “simple input-output models of deliberation effects [...] follows normative assumptions” is from Price (2009), “Citizens Deliberating Online: Theory and Some Evidence,” p. 41 of the chapter (pp. 37–58). Price, Nir and Cappella (2005) is a Public Opinion Quarterly article on pp. 179–212. Communication and Persuasion is by Hovland, Janis and Kelley (Yale University Press, 1953).
p. 513, Knowledge and Opinion Shift and Polarization
Published
“gains in issue-knowledge are modest at best” (Price, 2009, p. 15). [...] (Moscovici & Zavallon, 1969)
Correction
“gains in issue-knowledge are modest at best” (Price, 2009, p. 51). [...] (Moscovici & Zavalloni, 1969)
Basis The quoted phrase is on p. 51 of Price (2009), pp. 37–58. The reference list gives Moscovici and Zavalloni (1969).
p. 515, Data and Measures
Published
(e.g., see Karpowitz et al., 2012; Mendelberg, 2002; Mendelberg et al., 2013)
Correction
(e.g., see Karpowitz et al., 2012; Mendelberg, 2002; Mendelberg et al., 2014)
Basis See the correction to p. 509.
p. 515, Data and Measures
Published
The final number of participants included in this analysis was 264. The average small group size was 15.53 (SD = 2.95).
Correction
The 15 discussion groups included in this analysis had 233 participants. The average small group size was 15.53 (SD = 2.95).
Basis SI Tables 1–3 list 233 participants across groups A–O (SI Table 4, party: 231), and these group sizes have the mean and SD printed here. With 264 participants in 15 groups, the mean group size would be 17.6.
p. 518, Table 2
Published
Table 2, row “Mean (s.e.)”, column “Statements to an individual”: 1.73 (0.02).
Correction
Corrected Table 2. The mean for statements to an individual is 1.73 with s.e. 0.03.
Corrected Table 2. Counts and percentages as printed; means and standard errors recomputed from the printed counts (justification coded 0-3). Statements to an individual Statements to the group No justification 87 (10.60%) 475 (31.44%) Inferior justification 122 (14.86%) 416 (27.53%) Qualified justification 537 (65.40%) 601 (39.77%) Sophisticated justification 75 (9.14%) 19 (1.26%) N 821 1,511 Mean (s.e.) 1.73 (0.03) 1.11 (0.02) Basis From the printed counts, the standard error is 0.027, which rounds to 0.03.
p. 519, Table 3
Published
Table 3, column “Free trade”: “(Intercept)” 0.84∗ (0.44); all four models: “Num. participants” 264.
Correction
Corrected Table 3. The free-trade intercept, 0.84 (0.44), is not significant (z = 1.91, p = .056). No model can include more than the 233 participants in the 15 discussion groups; the exact number in each model cannot be recovered from the published materials.
Corrected Table 3: estimates as printed; the free-trade intercept is no longer starred, and the participant count is bounded by the 233 participants in the 15 discussion groups (SI Tables 1-3). * p < .05. Term Iraq Subsidies Free trade Protectionism (Intercept) 0.31 0.18 0.84 0.50 (0.40) (0.43) (0.44) (0.36) Liberal interactions 0.22* 0.34* 0.21* 0.32* (0.08) (0.13) (0.07) (0.06) Conservative interactions −0.07* −0.06* −0.10* −0.02 (0.03) (0.02) (0.03) (0.10) Dist. group attitude −0.04 0.01 −0.30* −0.37* (0.07) (0.12) (0.10) (0.08) Lib. interactions × know −0.14 0.09 −0.06 0.00 (0.60) (0.10) (0.11) (0.08) Con. interactions × know −0.15 0.05 0.05 0.07 (0.18) (0.09) (0.10) (0.11) Issue know change 0.20 −0.02 −0.06 −0.01 (0.23) (0.18) (0.19) (0.15) Republican −0.20 −0.06 −0.12 −0.19 (0.16) (0.19) (0.19) (0.15) Democrat 0.17 0.16 0.00 −0.47* (0.17) (0.19) (0.19) (0.16) Female −0.27* −0.05 −0.20 −0.02 (0.13) (0.15) (0.16) (0.13) High school 0.03 −0.08 −0.10 0.09 (0.22) (0.25) (0.26) (0.20) Some college −0.11 0.01 −0.24 0.29 (0.22) (0.25) (0.25) (0.20) College degree −0.29 −0.23 0.32 0.04 (0.22) (0.25) (0.26) (0.20) Age 0.01 0.00 −0.01 0.00 (0.00) (0.00) (0.00) (0.00) White −0.02 0.05 −0.17 −0.01 (0.20) (0.22) (0.23) (0.18) Num. participants ≤ 233 ≤ 233 ≤ 233 ≤ 233 Num. groups 15 15 15 15 Basis Even the most favorable rounding of the printed estimate and standard error (0.845 and 0.435) gives p = .052. On the participant count, see the correction to p. 515.
p. 520, Persuasive Interaction Predicts Opinion Change
Published
After deliberation, Democrats were more likely to adopt a more liberal position on the issue of protectionism (b = −.47, s.e. = .16, p < 0.05), but this was not observed in any of the other issues.
Correction
After deliberation, Democrats moved in the conservative direction on the issue of protectionism relative to respondents who were neither Democrats nor Republicans (b = −.47, s.e. = .16, p < 0.05), but this was not observed in any of the other issues.
Basis The dependent variable is T2 minus T1 opinion, and “a negative value indicates movement toward a more conservative opinion” (p. 518). Table 3 includes both Democrat and Republican indicators, so the reference group is all other respondents.
p. 520, Persuasive Interaction Predicts Opinion Change
Published
The effect sizes of interactions from more liberal peers were consistently larger than the effect of any other measure, while the effect size of interactions from more conservative peers were smaller than the effects of both initial group attitudes and the effects of interactions from more liberal peers.
Correction
The coefficients on interactions from more liberal peers were larger than those on issue knowledge change on all four issues, but distance from the initial group attitude had the larger coefficient on free trade (−0.30 against 0.21) and protectionism (−0.37 against 0.32). The coefficients on interactions from more conservative peers were smaller than those on interactions from more liberal peers on all four issues, and smaller than those on distance from the initial group attitude on the two issues where distance had a detectable effect.
Basis Table 3. Interactions from more liberal peers are significant on 4 of 4 issues and interactions from more conservative peers on 3, against 2 for distance from the group and 0 for issue knowledge change, so the claim that persuasion is the more consistent predictor stands. The raw coefficients are per justified statement received, per scale point of distance and per unit of knowledge change, so their sizes are not on a common scale.
p. 520, Persuasive Interaction Predicts Opinion Change
Published
Although the sample used in this study was recruited to match demographic variables, there were 7% more Democrats than Republicans and the number of Independents adopting liberal positions was twice as large as the number adopting conservative opinions. It is therefore not surprising that the effects of liberal interactions are more successful than conservative interactions— there are more liberals to respond to liberal discussion.
Correction
Although the sample used in this study was recruited to match demographic variables, the full sample had 7% more Democrats than Republicans and the number of Independents adopting liberal positions was twice as large as the number adopting conservative opinions. The 15 discussion groups analyzed here, however, had more Republicans (80) than Democrats (73), so the party composition of the analyzed groups does not explain why liberal interactions were more successful than conservative interactions.
Basis The 7% matches the full-sample demographics in SI section 1 (Democrat 36.28%, Republican 29.34%). SI Table 4, which covers the analyzed groups A–O, sums to 73 Democrats and 80 Republicans.
p. 520, Knowledge Change Is Not Predictive of Opinion Change
Published
(b = –.68, s.e. = 0.07, p = .03)
Correction
(b = −.68, s.e. = 0.31, p = .03)
Basis The printed s.e. cannot belong to this slope: b = −.68 with s.e. = 0.07 gives |t| = 9.7 and p far below .001. The standard error consistent with the printed coefficient and p-value is 0.31. See the correction to Figure 1.
p. 521, Figure 1, panel annotations
Published
Figure 1 annotations (b, s.e., p): Iraq (b=−0.68, s.e.=0.07, p=0.03); Multilateralism (b=0.1, s.e.=0.08, p=0.78); Preemptive war (b=−0.17, s.e.=0.08, p=0.61); NAFTA (b=0.51, s.e.=0.09, p=0.2); Subsidies (b=−0.29, s.e.=0.09, p=0.46); Free trade (b=−0.16, s.e.=0.08, p=0.65); Protectionism (b=−0.17, s.e.=0.08, p=0.61); Past econ. (b=0.25, s.e.=0.07, p=0.38); Future econ. (b=0.15, s.e.=0.06, p=0.6).
Correction
Corrected annotations. In every panel the printed s.e. is inconsistent with the printed slope and p-value. The slope standard errors consistent with the printed coefficients and p-values range from 0.28 to 0.40.
Figure 1 panel annotations. The printed s.e. values cannot belong to the printed slopes and p-values; the last column gives the slope standard error consistent with the printed coefficient and p-value. Panel b s.e. as printed p as printed p implied by b and printed s.e. s.e. implied by b and printed p Iraq −0.68 0.07 0.03 < 0.001 0.31 Multilateralism 0.10 0.08 0.78 0.213 0.36 Preemptive war −0.17 0.08 0.61 0.035 0.33 NAFTA 0.51 0.09 0.20 < 0.001 0.40 Subsidies −0.29 0.09 0.46 0.001 0.39 Free trade −0.16 0.08 0.65 0.047 0.35 Protectionism −0.17 0.08 0.61 0.035 0.33 Past econ. 0.25 0.07 0.38 < 0.001 0.28 Future econ. 0.15 0.06 0.60 0.013 0.29 Basis Computed from the printed b and p with 231 degrees of freedom (233 participants). Taking the printed p-values as correct, the conclusion that knowledge change predicts the amount of opinion change only for Iraq stands.
p. 522, Movement Toward Extremity Is Inconsistent
Published
On average, participants showed no change between T1 and T2 for 4.43 issues (s.e. = .11), movement toward a more extreme opinion on 2.40 issues (s.e. = 0.11), and movement toward the midpoint on 2.17 issues (s.e. = 0.09).
Correction
On average, participants showed no change between T1 and T2 for 4.70 issues, movement toward a more extreme opinion on 2.06 issues, and movement toward the midpoint on 2.24 issues.
Basis Averages implied by Table 4, which gives the percentage of participants in each category for each of the nine issues; the standard errors cannot be recomputed from the published table. The printed averages also sum to nine but cannot come from the same table: by Table 4, movement toward the midpoint is more common than movement toward a more extreme opinion.
p. 523, Discussion
Published
Persuasion has strong and consistent effects on opinion change after deliberation, and the effects of persuasion are larger and more consequential than both knowledge change and polarization.
Correction
Persuasion has strong and consistent effects on opinion change after deliberation; the effects of persuasion are larger than those of knowledge change and more consistent than those of polarization.
Basis See the correction to p. 520: distance from the initial group attitude, which p. 522 treats as one of the two measures of polarization, has the larger coefficient on free trade and protectionism.
p. 523, Discussion
Published
First, there were more liberals in the sample. [...] Liberals and conservatives are likely more responsive to information that aligns with their ideological perspective, with these effects magnified by the distribution of liberals in the sample.
Correction
First, the full sample had more Democrats than Republicans, although the 15 analyzed discussion groups did not (73 Democrats and 80 Republicans; SI Table 4). [...] Liberals and conservatives are likely more responsive to information that aligns with their ideological perspective.
Basis See the correction to p. 520.
p. 524, Discussion
Published
(Festinger, 1964; Howland et al., 1954)
Correction
(Festinger, 1964; Hovland et al., 1953)
Basis See the correction to p. 511.
pp. 526–527, References
Published
Fishkin, J. S. (2009). When the People speak: Deliberative democracy and public Consultation. New York, NY: University Press. Fishkin, J. S., He, B., Luskin, R. C., & SIU, a. (2010). [...] Howland, C. I., Kelley, H. H., & Janis, I. L. (1954). Communication and persuasion: Psychological studies of opinion change. New Haven, CT: Yale University Press. Mendelberg, T. (2002). [...] Political decision-making, deliberation and particivation [...] Mendelberg, T., Karpowitz, C. F., & Goedert, N. (2013). [...] Poole, M. S., McPhee, R. D., & Seibold, D. R. (1982). [...] Communications Managraphs, 49(1), 1–19. Reis, H. T. (2010). [...] In R. F. Baumesiter & E. J. Finkel (Eds.) [...]
Correction
Fishkin, J. S. (2009). When the People speak: Deliberative democracy and public Consultation. New York, NY: Oxford University Press.
Fishkin, J. S., He, B., Luskin, R. C., & Siu, A. (2010). [...]
Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion: Psychological studies of opinion change. New Haven, CT: Yale University Press.
Mendelberg, T. (2002). [...] Political decision-making, deliberation and participation [...]
Mendelberg, T., Karpowitz, C. F., & Goedert, N. (2014). [...]
Poole, M. S., McPhee, R. D., & Seibold, D. R. (1982). [...] Communications Monographs, 49(1), 1–19.
Reis, H. T. (2010). [...] In R. F. Baumeister & E. J. Finkel (Eds.) [...]
Supplementary information
SI p. 1, section 1 (Participant Demographics)
Published
Section 1 (“Participant Demographics”): “Gender”: “Male” 45.64%, “Female” 54.36%; “Education”: “Less than high school” 15.44%, “High school” 29.19%, “Some college” 29.53%, “College degree and more” 25.84%; “Political party”: “Republican” 29.34%, “Democrat” 36.28%, “Independent” 27.13%, “Other party” 6.62%.
Correction
Demographics of the 233 participants in the 15 analyzed discussion groups, from SI Tables 1–4. 56.2% are men, and Republicans outnumber Democrats (80 to 73).
Participant demographics in the 15 analyzed discussion groups, from SI Tables 1-4 (N = 233; party N = 231). Characteristic Category 15 analyzed groups Gender Male 56.2% Gender Female 43.8% Age 18-30 21.5% Age 31-40 18.5% Age 41-50 17.2% Age 51-60 18.9% Age 60 and older 24.0% Education Less than high school 14.2% Education High school 26.6% Education Some college 27.9% Education Bachelors and beyond 31.3% Political party Democrat 31.6% Political party Republican 34.6% Political party Independent 27.7% Political party Other 6.1% Basis The printed party and education percentages cannot be shares of any of the samples the article describes (284 deliberators, 264 participants, or the 233 participants in SI Tables 1–3): no whole-number count out of those totals rounds to them.
SI p. 2, section 2 (Randomization check) and Table 2
Published
An OLS model for discussion group with demographic traits as predictors was then run. No significant differences were observed, though the model itself was significant. [...] χ2 (df=840) =831.1, p=.58 (continuous variable used)
Correction
Gender, education and party do not differ significantly across discussion groups (Tables 1, 3 and 4). Age does: for the age groups in Table 2, χ2 (df=56) = 81.6, p = .014. Group L has 10 members aged 60 or older out of 19; group O has 0. [...] Table 2: χ2 (df=56) = 81.6, p = .014
Basis The printed age test treats each year of age as a category: df = 840 = 14 × 60 implies 61 ages in 15 groups, 915 cells for about 233 participants (0.25 per cell), too sparse for a χ2 test. A simulation-based p-value for the age-group table is .014. A regression with the discussion-group letter as the outcome does not test balance, and its significance contradicts the statement that no differences were observed. The gender, education and party tests reproduce from Tables 1, 3 and 4.