error associated with the estimate, 1.5% (see Table 2, column 3). Pooled together, the
experiments indicate that the group receiving the negative campaign message were 0.6
percentage points more likely to vote than subjects exposed to the positive campaign
message (see Table 2, column 4). Thus, the two experiments indicate that subjects were
unlikely to be demobilized by exposure to a negative message from the campaign rather
than a positive one.
Table 2: Voter Turnout by Randomly Assigned Message Tone
Harvested
Purchased
Pooled
Turnout in Negative
Group
55.7%
[911]
39.3%
[2211]
Turnout in Positive
Group
52.5%
[910]
39.8%
[2174]
Difference
+3.1%
(2.3)
-0.5%
(1.5)
+0.6%
(1.3)
p-value
(one-tailed)
0.90 0.37 0.67
Numbers in brackets represent N.
Numbers in parentheses represent standard errors.
The picture is much the same for voter preference and attitudes about the
candidates. The positive message group in the harvested sample was 2.3 percentage
points more likely to have voted for Kerry (see Table 3 column 2), but there was no
difference whatsoever among the purchased group (see Table 3, column 3). Pooled
together, positive messages outperformed negative messages by less than a percentage
point, 0.8, and one-third the size of the standard error, 2.5 (see Table 3, column 4). Thus,
it appears unlikely that negative messages are substantially more effective than positive
messages at swaying voters.
18
Convention