3.4 The present research
3.4.2 Experiment 1: setting up a selective exposure paradigm
3.4.2.4 Results
After removing any participants who did not complete the experiment or who failed
‘attention checks’, we are left with 195 subjects who completed all questions on in- come inequality (of which 28 were ‘moderates’ - expressing no initial opinion), 172 who
completed all questions on the minimum wage (of which 32 were moderates), 168 who completed all questions on the death penalty (of which 50 were moderates), and 161
who completed all questions on gun control (of which 21 were moderates.) 1
Recall that participants were randomly allocated to one of three conditions: a control
(‘opinions’) condition, or one of two treatment conditions (‘knowledge’ or ‘knowledge and answers’.) The below table shows the breakdown of participants in each condition
and some key demographic data.2
Sample size Gender (%F) Age (mean) Income (mean $) Opinion 66 48% 39.5 41,841 Knowledge 50 58% 43.4 45,260 Answers 80 50% 35.3 51,831
Table 3.1: Demographic characteristics, experiment 1
1
For each topic, we include in the analysis all participants who completed all the questions forthat topic- even if they did not complete the whole experiment - therefore essentially looking at four different datasets, one for each of the four topics.
2
Note that the size of the three conditions is somewhat uneven - perhaps due to the relatively small sample size. This is worth bearing in mind when interpreting the results.
A one-way ANOVA finds no significant difference in the mean ‘bias score’ (no. of sup- porting arguments - no. of conflicting arguments selected) between the three condi-
tions, for any of the four topics. (income inequality: F(2,164)=0.27, p=0.76; minimum wage: F(2,137)=0.25, p=0.78; gun control: F(2,137)=0.59, p=0.56; death penalty:
F(2,115)=0.69, p=0.51). This suggests that the effect of our intervention was not as hypothesised: the framing of the experiment did not affect the arguments people chose
to read.
We can also look at the average bias scores for each of the four topics for each group3, and ask whether they differ significantly from zero (i.e. do people choose significantly more supporting/conflicting arguments?)
When we look at the average bias scores across all four topics for each group, we find that the average bias score for the opinions condition is significant (mean = 0.53, t=2.68,
p < 0.05 - see table 3.2 below), but not for either of the other two conditions. We therefore find some evidence of a selective exposure effect when people are asked about
their opinions on political issues, but no evidence of such an effect when people are asked about their knowledge on those same issues. This might be interpreted as support for
our initial hypothesis, but given the non-significant ANOVA results, this seems weak evidence at best.
The pattern is also less clear when we look at the breakdown of bias scores by topics. We dont see the same pattern holding for peoples information choices for the separate topics,
but only when we average choices across them all. Looking across all three conditions, we find that the bias score only significantly differs from zero for the topic of income
inequality (mean bias score = 0.57, t=3.57, p < 0.001). Within the topic of income inequality, the bias score is significant for those in the opinions condition (mean = 0.63,
t=2.34,p <0.05), and the answers condition (mean = 0.66, t=2.56,p <0.05), but not in the knowledge condition (mean = 0.37, t=1.14,p= 0.2627.) That is, we find evidence of selective exposure in the opinion and answers conditions, but not in the knowledge condition. This is somewhat surprising, given that the knowledge and answers conditions
were identical except for the fact that those in the latter group expected to be given factual answers to some of the questions at the end, which we hypothesized would make
3
Here, we take the ‘bias score’ of each participant for each topic, and average the four values to get an ‘average bias score’ for each participant. These average bias scores are then averaged across individuals in each condition, to get an overall bias score for each condition.
people more likely to seek balanced information, not less. For the other three topics, bias scores do not significantly differ from zero for any groups: there is no evidence that
people display any consistent preference for supportive over conflicting information.
One way to interpret this might be that the manipulation does subtly influence selective
exposure, ut the extent to which selective exposure occurs also depends on certain fea- tures of the topic/issue - so that when we look across several topics the influence of the
framing becomes apparent, but for individual topics the influence of the framing may be outweighed by factors specific to the topic.
Finally, we note that when we look at average bias scores across all three conditions, we only find a significant bias score for one of the four topics, and when averaged across all
four topics, the mean bias score (0.23) does not significantly differ from zero (t=1.68, p=0.095). This suggests that, though selective exposure seems to occur under certain
conditions, this study provides little evidence that it holds as a broad tendency across a wide range of scenarios.
Income Inequality Minimum Wage Gun Control Death Penalty Average - all topics Opinion 0.63* 0.22 0.34 -0.12 0.53* Knowledge 0.37 0.00 0.29 0.38 0.18 Answers 0.66* -0.07 -0.03 -0.11 0.07 Average 0.57*** 0.05 0.20 0.05 0.23 *** p <0.01,**p <0.01,*p <0.0.5
Table 3.2: Bias scores by condition and topic, experiment 1