The Minimal Persuasive Effects of Campaign Contact in General Elections: Evidence from 49 Field Experiments
American Political Science Review, 2017 · paper · Political science
The text provided is only an online appendix of supplementary figures and tables, including analyses of partisanship interactions, independents' persuadability, and recall of campaign contact. It does not state the main findings of the 49 field experiments.
Checked: PDF from static.cambridge.org, fetched 5 Oct 2026. Page numbers are those of that PDF.
7 findings
Substance B.3.2: "-0.0105 (0.0103) for the ITT. For the CACE, dividing by 0.14, this is 0.075 (0.074)." A negative ITT divided by 0.14 gives -0.075, so the printed CACE has the wrong sign.
- p. 8 B.3.2 Phone Call
-0.0105 (0.0103) for the ITT. For the CACE, dividing by 0.14, this is 0.075 (0.074)
- p. 8 B.3.2 Phone Call
-0.0105 (0.0103) for the ITT. For the CACE, dividing by 0.14, this is 0.075 (0.074).
Judge's note: Page 8 image: the ITT carries a minus sign and the CACE 0.075 does not. On page 7, B.3.1 applies the same procedure and keeps the sign (-0.0188/0.2 = -0.094). The SE 0.074 is consistent.
Second check: another model agreed
Report a problem with this finding
Substance B.7, Cubbison (2015): "prtesti 180 .4556 1223 .4557 yields an estimate of 0.0001 (0.0142)". The SE of a difference of two proportions with these inputs is about 0.040.
- p. 11 B.7 Cubbison (2015)
yields an estimate of 0.0001 (0.0142)
Only the authors can say what is right here.
Judge's note: Page 11 image. The difference SE is sqrt(.4556*.5444/180 + .4557*.5443/1233), about 0.0397, and the pooled SE is similar. The printed 0.0142 equals the SE of the treatment-group proportion alone (sqrt(.4557*.5443/1233) = 0.01418). It cannot come from the stated prtesti test. The same procedure in B.6 reproduces its printed SE (0.038).
Second check: another model agreed
Report a problem with this finding
Substance B.9.1, Gerber (2004) Study 1: "the pooled estimate and SE is 0.033 (0.067). Dividing by two ... the overall pooled estimate is 0.016 (0.036)." Half of 0.067 is about 0.034.
- p. 13 B.9.1 Study 1
Dividing by two since this is reported in vote margin terms, the overall pooled estimate is 0.016 (0.036)
- p. 13 B.9.1 Study 1, treatment effect bullet
the pooled estimate and SE is 0.033 (0.067). Dividing by two since this is reported in vote margin terms, the overall pooled estimate is 0.016 (0.036).
- p. 13 B.9.1 Study 1
the overall pooled estimate is 0.016 (0.036)
- p. 13 B.9.1
Dividing by two since this is reported in vote margin terms, the overall pooled estimate is 0.016 (0.036).
Only the authors can say what is right here.
Judge's note: Page 13 image. I redid the inverse-variance pool of 0.08 (0.11), -0.12 (0.12) and 0.13 (0.12): estimate 0.033, SE 0.0672. Halving gives 0.0165 (0.0336). The estimate matches but the SE does not. Study 2 on page 14 halves correctly (0.075 to 0.037).
Second check: another model agreed
Report a problem with this finding
Substance B.20, Shaw and Gimpel (2012): "The surveys were taken January 10-20, eight months before the election, so we use -235 days." B.10 counts mid-January to the November 7 general election in the same race as 296 days.
- p. 26 B.20 Shaw and Gimpel (2012)
The surveys were taken January 10-20, eight months before the election, so we use -235 days.
- p. 16 B.10 Gerber et al. (2011)
running in mid-January with a November 7 general election, this experiment occurred 296 days before the election.
- p. 16 B.10 Gerber et al. (2011), Days after election bullet
running in mid-January with a November 7 general election, this experiment occurred 296 days
- p. 16 B.10 Gerber et al. (2011)
this experiment occurred 296 days before the election
Only the authors can say what is right here.
Judge's note: Pages 26 and 16 images. Both entries describe the Texas governor general election with the same 9.2-point margin and the same competitiveness wording. Counting from January 10–20 to November 7 gives about 291–301 days, roughly ten months. Neither 'eight months' nor -235 follows from the stated dates. Rests on outside knowledge only to the extent that B.20 itself does not restate the election date; B.10 gives it as November 7.
Second check: another model agreed
Report a problem with this finding
Substance B.10.1: the effect one week later is given as "-0.0017 (0.0142)", but the quotation offered as its source says "receded to 0.17 percentage points (SE = 1.42)", which is positive.
- p. 17 B.10.1 One week later
later is -0.0017 (0.0142): “a week later the effects of these ads have receded to 0.17 percentage points (SE = 1.42).”
Only the authors can say what is right here.
Judge's note: Page 17 image: the minus sign is visible on the coded value, and the quoted source gives a positive 0.17 points. Nothing in the entry explains the sign difference. The original article's table would settle which sign is right.
Second check: another model agreed
Report a problem with this finding
Wording B.2, Arceneaux and Kolodny: one bullet says "the survey began two days after the election ... We use 7." The Mode of measurement bullet then says "We assume the survey was taken the day after the election."
- p. 6 B.2, Days after election
We use 7.
- p. 6 B.2, Mode of measurement
We assume the survey was taken the day after the election.
- p. 6 B.2, Days after election bullet
the survey began two days after the election and finished interviewing within three weeks. We use 7.
Only the authors can say what is right here.
Judge's note: Page 6 image: both bullets are in the same entry. The 14-day treatment-to-survey figure (7 + 7) only works with the 7-day value, so the 'day after' sentence contradicts the coding as printed. Only the authors can say which value was used.
Second check: another model agreed
Report a problem with this finding
Wording FL experiment: "Canvassing took place from . An initial post-treatment survey took place from ." Both date ranges are blank.
- p. 81 FL Experiment, 2016, intro
Canvassing took place from . An initial post-treatment survey took place from .
- p. 90 Other experiments' intros
Canvassing took place from 9/30/16-10/15/16
- p. 81 FL Experiment intro
The election was held on 11/8/16.
Only the authors can say what is right here.
Judge's note: The p81 page image shows the blanks, so this is not an extraction artefact. Other sections (e.g. p90) give dates. Only the authors can supply the missing dates.
Second check: another model agreed
Report a problem with this finding
23 minor (typos and the like; most checked by the judge only)
Wording B.6, Cardy (2005): the text says "The control group had 192 individuals with 64% voting for the candidate", but the command printed is "prtesti 192 .65 805 .673 yields 0.023 (0.038)".
- p. 10 B.6 Cardy (2005), treatment effect
192 individuals with 64% voting for the candidate
- p. 10 B.6 Cardy (2005), treatment effect
prtesti 192 .65 805 .673 yields 0.023 (0.038)
- p. 10 B.6 Cardy (2005), treatment effect
The control group had 192 individuals with 64% voting for the candidate
- p. 10 B.6 Cardy (2005), same bullet
prtesti 192 .65 805
Only the authors can say what is right here.
Judge's note: Page 10 image. The estimate 0.023 and SE 0.038 reproduce with .65, not with .64. A proportion near 0.645 could be rounded either way, but as printed one control share is given two ways. Only the authors can say which is right.
Second check: another model agreed
Report a problem with this finding
Wording B.11: the text gives "about 9.5 percentage points (= 0.170 −0.076)" and "because ITTD is 9.5 percentage points here", but 0.170−0.076 = 0.094, and the SE of 18.1 = 1.7/0.094.
- p. 18 B.11
about 9.5 percentage points (= 0.170 −0.076)
- p. 18 B.11
because ITTD is 9.5 percentage points here
Judge's note: Page 18 image. The 9.5 sits inside a quotation from the source and could reflect unrounded inputs. The appendix's own statement that ITT_D is 9.5, however, disagrees with the fraction it uses for the SE (0.094). With 0.095 the SE would be 17.9. A minor consistency point.
Second check: another model agreed
Report a problem with this finding
Wording B.19, Sadin (2014): "The survey was conducted about 7 weeks before the election, so we use 50." Other pre-election surveys are entered as negative values (e.g. -13, -1, -235).
- p. 25 B.19 Sadin (2014), Days after election bullet
The survey was conducted about 7 weeks before the election, so we use 50.
Only the authors can say what is right here.
Judge's note: Page 25 image. Elsewhere the appendix enters pre-election timing with a minus sign (B.3 -13 on page 7; B.5.2 -1 on page 10; B.20 -235 on page 26). Some other prose entries (e.g. B.8.3, B.25) also omit the sign, so the dataset may hold -50. As printed, the value is ambiguous.
Second check: another model agreed
Report a problem with this finding
Wording B.25, Green (2012b): "Incumbent for President and Senate, varies for Senate." The seats are "President, Senate, and State House", so the last item is evidently meant to be State House.
- p. 29 B.25 Green (2012b), Incumbency
Incumbent for President and Senate, varies for Senate.
- p. 29 B.25 Green (2012b), Seat
President, Senate, and State House.
Judge's note: Page 29 image: Senate appears twice with conflicting codes, and State House gets no incumbency code.
Second check: another model agreed
Report a problem with this finding
Wording Appendix D overview says Working America "would randomly assign half to a treatment group ... and half to a placebo group", but the FL canvassed counts are Treatment 292 vs Placebo 449 (39% treatment), and OH1 has 310 vs 130.
- p. 41 Appendix D Overview, step 3
would randomly assign half to a treatment group
- p. 83 FL experiment, counts by condition
Treatment 292 ... Placebo 449
- p. 57 OH Experiment 1, counts by condition
Treatment 310 ... Placebo 130
Only the authors can say what is right here.
Judge's note: Checked the overview step 3 on p41 and the count tables on p83 (FL) and p57 (OH1). The FL balance regression (Table 44, p82) has intercept 0.39 for the treatment indicator, which matches about 39% treatment among those canvassed. OH1 used two treatment scripts that were merged (p57), which plausibly explains its roughly 2:1 split. Nothing explains FL. The counts are of people actually canvassed, not of people assigned, so differential contact could contribute. Only the authors can say what the FL assignment ratio was. Severity is S3 because this is a wording overstatement of the design.
Second check: another model agreed
Report a problem with this finding
Wording The repeated note says "t0_pid is the standard 7-point party ID variable", but the reported means include 0.5 (PA), 0.26 (OH1), 0.25 (OH2), 0.75 (FL) and 0.63 (MO), while WA reports 4.1.
- p. 42 Representativeness table note (repeated for each experiment)
t0_pid is the standard 7-point party ID variable, with higher values for stronger Democrats.
- p. 64 OH Experiment 2 representativeness table
t0_pid
- p. 41 PA Experimental Universe table
t0_pid 0.5 0.47
- p. 42 Representativeness caption (repeated for every experiment)
t0_pid is the standard 7-point party ID variable
Only the authors can say what is right here.
Judge's note: The page images confirm PA t0_pid 0.5/0.47 (p41), WA 4.1/4 (p49) and OH2 0.25/0.41 (p64). FL shows 0.75 (p82), and the text layer gives OH1 0.26, NC 1.4 and MO 0.63. Means below 1 cannot occur on a 1–7 scale, so the coding evidently differs across experiments (for example, centred). The note is identical under every table and never describes this. Only the authors can state the coding used.
Second check: another model agreed
Report a problem with this finding
Wording OH Experiment 2: "A second follow-up post-treatment survey took place from 11/8/16. The election was held on 11/8/16." The range has no end date.
- p. 64 OH Experiment 2 intro
A second follow-up post-treatment survey took place from 11/8/16.
- p. 64 OH Experiment 2 intro
A second follow-up post-treatment survey took place from 11/8/16. The election was held on 11/8/16.
Only the authors can say what is right here.
Judge's note: The p64 image shows only a start date after "from", and it is the same as election day. Every other section gives a full range. Only the authors can supply the intended dates.
Second check: another model agreed
Report a problem with this finding
Wording "As part of these canvasses, North Carolina also distributed literature" names the state as the actor, but the paragraph identifies Working America as the canvasser.
- p. 72 NC Experiment intro
As part of these canvasses, North Carolina also distributed literature
Judge's note: The p72 image shows the wording. The preceding sentence says Working America canvassed, so "North Carolina" is evidently a slip for Working America (or for the NC program).
Second check: another model agreed
Report a problem with this finding
Wording OH DiD wave sizes '(n=3,545)... (n=1,823)... (n=1,621)... (n=1,649)... (n=1,328)' (p.114) sum to 9,966, but the OH DiD results table (p.121) prints 'N obs 9906' for the Everyone column.
- p. 114 OH Difference-in-Differences, wave sizes
The first wave was conducted from 7-19 October (n=3,545)
- p. 121 OH DiD Results table
9906
- p. 114 OH Difference-in-Differences, wave description
(n=3,545). The second wave was conducted from 20-29 October (n=1,823)
- p. 121 OH DiD results table, N obs
N obs 9906
Only the authors can say what is right here.
Judge's note: The p.114 image confirms the five wave sizes. The p.121 image shows N obs 9906 and N groups 3545 in the OH DiD Results table, which sits in the same OH Difference-in-Differences section. The NC DiD wave sizes (p.113) sum exactly to its printed N obs of 18,290. I found no statement of dropped observations or missing outcomes for OH. A transposition of 9966 is plausible, but 60 rows lost to missing outcomes cannot be ruled out, so the authors need to say which figure is right.
Second check: another model agreed
Report a problem with this finding
Typo The Figure OA4 caption is copied word for word from Table OA1. It says "This table shows the relative infrequency of personal contact" and "We present mean recall rates and the standard error of the mean", but the bar chart has no error bars.
- p. 4 Figure OA4 caption
We present mean recall rates and the standard error of the mean for multiple types of campaign contact.
- p. 4 Figure OA4 caption
This table shows the
- p. 4 Figure OA4 caption
This table shows the relative infrequency of personal contact
Judge's note: Page 4 image: Figure OA4 is a bar chart with no SE bars, and its caption is identical to the Table OA1 caption beneath it. The data match. Only the caption wording is wrong for a figure.
Second check: none (judge only)
Report a problem with this finding
Typo The Figure OA4 y-axis is labelled "Mean Recall Percentage", but the tick marks run 0.00–0.75, which are proportions.
- p. 4 Figure OA4 y-axis
Mean Recall Percentage
Judge's note: Page 4 image: the ticks are 0.00, 0.25, 0.50 and 0.75, matching the Table OA1 means (e.g. 0.243). The scale is proportion while the label says percentage. Loose wording, and the meaning can be recovered.
Second check: none (judge only)
Report a problem with this finding
Typo B.7, Cubbison (2015): the text sums the groups as "129 + 119 + 185 + 418 + 382 = 1, 233 subjects", but the command uses "prtesti 180 .4556 1223 .4557".
- p. 11 B.7 Cubbison (2015)
129 + 119 + 185 + 418 + 382 = 1, 233 subjects
- p. 11 B.7 Cubbison (2015)
prtesti 180 .4556 1223 .4557
- p. 11 B.7 Cubbison (2015), treatment effect
= 1, 233 subjects
Judge's note: Page 11 image. The sum is 1233, and the proportion .4557 equals 562/1233 truncated (0.45580), not 562/1223 (0.4595). So 1223 looks like a transposition typo in the printed command. The 45.57% is a truncation, not an error.
Second check: none (judge only)
Report a problem with this finding
Typo B.9.4: "Competitiveiness. No." is a misspelling of "Competitiveness".
- p. 15 B.9.4
Competitiveiness.
- p. 15 B.9.4 Study 4, last bullet
Competitiveiness. No.
Judge's note: Page 15 image: the misspelling is in the last bullet of B.9.4.
Second check: none (judge only)
Report a problem with this finding
Typo Misspellings in the appendix text: "7-pint scales", "individals", "survey responded identified" (repeated table note), "apprximately" (p106), "everyone was is never canvassed" (p114), "parantheses" (p114), "Competitiveiness" (p15).
- p. 42 Representativeness table note
Favorability variables are 7-pint scales
- p. 42 Representativeness table note
the number of individals at each stage
- p. 106 NC GOTV results text
is apprximately 43% more
- p. 114 OH DiD Tests of Trends
compare those individuals to everyone was is never canvassed
Judge's note: Each one is visible in the page images: the table note on p42, which also appears under later tables (p82); p106; p114; and p15. These are cosmetic typos, not extraction artefacts.
Second check: none (judge only)
Report a problem with this finding
Typo The reproduced script footer reads "1 Washginton LD 30 Experiment Training Rap" (Washington misspelled).
- p. 52
1 Washginton LD 30 Experiment Training Rap
Judge's note: Visible in the p52 crop. It is in the footer of the reproduced partner script, and it is cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo "Because there was no statistically significant differences" has a subject-verb agreement error (should be "were").
- p. 57 OH Experiment 1 Description of Treatment
there was no statistically significant differences
Judge's note: Visible on the p57 page image.
Second check: none (judge only)
Report a problem with this finding
Typo NC outcome item: "Wake Country Judge Michael Morgan" should read "Wake County".
- p. 80 NC Outcome Measures, Supreme Court item 2
Wake Country Judge Michael Morgan
Judge's note: Visible in the p80 crop. It is a cosmetic misspelling, possibly carried over from the instrument.
Second check: none (judge only)
Report a problem with this finding
Typo The Florida placebo script ("WA Florida Latino Partisanship Study Placebo Rap") says "better understand how to reach Missourians on issues of importance".
- p. 84
better understand how to reach Missourians on issues of
Judge's note: The p84 image confirms this is under the Florida study header. It is a reproduced partner script, apparently copied from the MO script. The error is in the reproduced material and is cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo The MO general-election script asks "who do you think you will be supporting in the Republican primary, Eric Greitens or Chris Koster?" although the same page frames this as the Governor's race.
- p. 95
who do you think you will be supporting in the Republican primary
Judge's note: Seen on the p95 image. The same script says "supporting Chris Koster in the Governor's race" and has the branch "[IF UNDECIDED OR GREITENS]", so "Republican primary" is inconsistent within the reproduced partner script. It is a verbatim script artefact and cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo The MO script says "supporting Koster for Governor because of her strong track record".
- p. 95
Koster for Governor because of her strong track record
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that Chris Koster is male. The p95 image shows "her". Nothing in the files states Koster's gender, and the text is a verbatim reproduction of the partner script.
Second check: none (judge only)
Report a problem with this finding
Typo The prose prints "515752 people randomly assigned" and "89271 people randomly assigned", while the adjacent tables print 515,752 and 89,271.
- p. 96 NC GOTV Experimental Universe
515752 people randomly assigned
- p. 106 MO GOTV Experimental Universe
89271 people randomly assigned
Judge's note: The p96 crop shows 515752 in the prose, and the text layer has 515,752 and 89,271 in the tables. The values agree; only the number format differs. This is cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Copy-editing errors in the GOTV/DiD appendix sections: 'compare those individuals to everyone was is never canvassed' (p.114; should be 'everyone who is/was never canvassed'), 'apprximately 43% more' (p.106), and 'this indicator is set to 1 and then is then always coded' (p.110; redundant 'then').
- p. 110 Identification Strategy
this indicator is set to 1 and then is then always coded as canvassed thereafter
- p. 114 OH DiD Tests of Trends
everyone was is never canvassed
- p. 106 NC GOTV results
apprximately 43% more
- p. 114 OH DiD, Tests of Trends Assumption
compare those individuals to everyone was is never canvassed
Judge's note: Checked on the crops. p.114 prints 'everyone was is never canvassed', while the parallel NC sentence on p.111 reads 'everyone who was never canvassed'. p.106 prints 'apprximately'. p.110 prints 'set to 1 and then is then always coded'. The 'parantheses' misspelling is handled separately under C034.
Second check: none (judge only)
Report a problem with this finding
Typo 'parentheses' is misspelled as 'parantheses' in 'Standard errors are reported in parantheses' (p.111) and 'Cluster-robust standard errors are reported in parantheses' (p.114, and again on p.121).
- p. 111 NC DiD, Tests of Trends Assumption
Standard errors are reported in parantheses
- p. 114 Results
Cluster-robust standard errors are reported in parantheses.
Judge's note: The misspelling appears in the C034-a and C034-b crops and on the full page 121 image (OH DiD Results paragraph).
Second check: none (judge only)