Backlash to Anti-Hate-Crime Policy: Evidence from an Online RCT

See experimental design section.

2026-08-10

2026-10-01

Exposure to anti-hate-crime legislation
Moral Conviction
Hate Crime Attitudes

Updated on 24/09/2026. All changes made in this update are marked by *** at the beginning and end of each revised passage.



Exposure to anti-hate crime legislation is a binary: participants in the control group are asked to read a dictionary and scientific definition of hate crime. Participants in the treatment group additionally read a text explaining UK Anti-Hate-Crime legislation.

Moral conviction is constructed from 3 questions based on Skitka et al. (2021):

A: The government and public institutions have a responsibility to reduce unequal treatment and unequal outcomes resulting from discrimination affecting some groups in the UK.
1 = strongly disagree
2 = disagree
3 = neither agree nor disagree
4 = agree
5 = strongly agree

B: Thinking about your feelings about this issue: To what extent are your feelings about this issue a reflection of your core moral values and convictions?
1 = not at all
2 = slightly
3 = moderately
4 = much
5 = very much

C: To what extent are your feelings about this issue deeply connected to your beliefs about “right” and “wrong”?
1 = not at all
2 = slightly
3 = moderately
4 = much
5 = very much


The dependent variable 'hate crime attitudes' is measured using the Universal Hate Crime Beliefs Scale (HCBS-U) (Bender et al., 2025) and an Implicit Association Test (Greenwald et al., 1998).

Hypothesis: (Backlash Effect)
The effect of exposure to anti-hate-crime laws on Hate Crime Attitudes will depend on policy alignment of moral conviction and may eventually reverse with strong policy adverse moral conviction.



***Update 24/09/2026***

Preliminary data suggested that it may be useful to include an additional measure of self-reported attitudes toward hate crime. The following item was therefore added at the very end of the post-test questionnaire. As it appears only after all other measures in the post-test questionnaire, its inclusion does not affect any of the previously specified measurements in the survey experiment.

Please indicate how reprehensible you would find it if someone behaved as follows.

D: When someone physically or verbally assaults a person because of their migration background or nationality.

1 = Not reprehensible at all
2
3
4 = Neither
5
6
7 = Very reprehensible

E: When someone physically or verbally assaults a person because of their sexual orientation.

1 = Not reprehensible at all
2
3
4 = Neither
5
6
7 = Very reprehensible

F: When someone physically or verbally assaults a person because of their skin color, ethnicity or religion.

1 = Not reprehensible at all
2
3
4 = Neither
5
6
7 = Very reprehensible

********************************************

Updated on 24/09/2026. All changes made in this update are marked by *** at the beginning and end of each revised passage.


***Update 24/09/2026***

Preliminary data suggested that it may be informative to include a measure of institutional trust. The following item was therefore added at the very end of the post-treatment questionnaire. As it appears only after all other outcome measures, its inclusion does not affect any of the previously specified measurements in the experiment.

Please indicate on a scale from 0 to 10 how much you personally trust each of the following institutions.

0 = No trust at all
10 = Complete trust

1. The UK Parliament

2. The legal system

3. The police

4. Politicians

5. Political parties

Response scale for each item:
0 = No trust at all
1
2
3
4
5
6
7
8
9
10 = Complete trust

******************************************

The study uses an between-subjects randomised online survey experiment. After completing a pre-test questionnaire on demographics and moral convictions, participants are randomly assigned to a treatment or control group and read a short informational text about hate crime. The treatment group reads an additional paragraph on anti-hate-crime legislation. Participants then complete an Implicit Association Test (IAT) and a questionnaire measuring approval of statements related to hate crime.

Participants complete the same otree app and on the information page are randomly shown either the treatment text or the control text. The assigned condition is captured as binary variable.

Treatment is assigned at the individual level.

No

No clusters

Updated on 24/09/2026. All changes made in this update are marked by *** at the beginning and end of each revised passage. *** Update 24/09/2026 *** Data collection was stopped after 168 observations because quota sampling based on political affiliation did not generate sufficient variation in the moral conviction measure. The recruitment strategy was therefore revised, and the two-phase design outlined below is no longer applicable. Additional participants will instead be recruited using a custom screening procedure based on the directional component of the moral conviction measure (Question A). A detailed description of the revised recruitment strategy and the resulting changes to the study design is provided in the attached document, “Revised Prolific Recruitment with Screening.” ****************************** *** Two-phase strategy abandoned on 24/09/2026 *** A maximum of 504 participants will be recruited through Prolific using its quota-sampling based on political affiliation (168 left- 168 centre - 168 right). The sample collection consists of two phases. Phase 1: 336 participants are recruited using Prolifics quota-sampling based on political affiliation: 168 left-leaning and 168 right-leaning. I expect collecting only left- and right-leaning political affilation to increase variation in the moral conviction moderator. Moral Conviction is here treated as a binary. If Phase 1 provides no support for Hypothesis 2, collecting centre-leaning participants is unlikely to be informative. I proceed to Phase 2 if the interaction is in the expected direction and the one-sided p-value is below 0.05 for at least one of the two hate-crime-attitude outcomes. Phase 2: 168 participants are recruited using Prolifics quota-sampling based on political affiliation: 168 centre-leaning. ***********************************************************************

~50% of participants in treatment group / ~50% in control group

Updated on 24/09/2026. All changes made in this update are marked by *** at the beginning and end of each revised passage. *** Two phase strategy abandoned on 24/09/2026 *** Phase 1: With N = 336, 80% power and one-sided α=.05, the minimum detectable difference in treatment effects between low- and high-moral-conviction groups is approximately 0.54 SD. ********************************************************************** Total sample: With N = 504, 80% power, and two-sided α = .05, the minimum detectable standardized treatment * moral-conviction interaction is approximately 0.25 SD per one-SD increase in moral conviction.