0
$\begingroup$

I have conducted a survey where participants are shown 8 different advertisements: 4 of the ads attempt to evoke the feeling of guilt, 4 others attempt to evoke the feeling of shame. After seeing each advertisement, I asked them to rate to what extend they feel guilt and shame on a 7-scale measure (1=not at all; 7=extremely). I have collected 70 respondents. I have checked that my data are not normally distributed because the advertisements are intended to evoke either guilt or shame. My research objective is to make sure that 4 of the guilt-inducing ads really evoke high feeling of guilt and low feeling of shame, and vice versa for the shame-inducing ads. In other words, an individual is assumed to give a significantly higher score of guilt for guilt-inducing advertisement compared to the shame-score. So, the guilt and shame score for each advertisement is significantly different.

My questions: 1. Is using a paired test correct? 2. Is it correct to use the Sign test as a non-parametric test to compare the guilt score and shame score for each advertisement?

I have actually tried both paired t-test and Sign test but both showed only 1 out of 8 advertisements have significantly different guilt and shame score.

Please let me know if I am on the right track. Thank you.

Improve this question
$\endgroup$
1
  • $\begingroup$ If by "I have checked that my data are not normally distributed" you mean by looking at the data in some fashion, this is pointless; even if you regard it as numeric, the response is integer between 1 and 7 - so right at the questionnaire design stage you know it cannot possibly be normally distributed. $\endgroup$
    – Glen_b
    Commented Mar 20, 2020 at 9:06

1 Answer 1

Reset to default
0
$\begingroup$

I spent some time on this and i have to say it was a bit hard understanding the research question and data structure. So I'm shooting a bit in the dark here.

Based on this line:

"In other words, an individual is assumed to give a significantly higher score of guilt for guilt-inducing advertisement compared to the shame-score."

Shame ads are not needed, and should have been a neutral "control ad", but here goes :)

First i would like to understand your data. I guess it looks something like this:

name  measured_guilt_ad_1  measured_shame_ad_1  measured_guilt_ad2  measured_shame ad_2 ...
Pete  1                    7                    5                   2
.
.
.

In that case I would take a mean of all the "measured guilt" and "measured shame" for guilt ads and shame ads for each participant, and use a paired t-test alternative a "Wilcoxon signed-rank test" rather then a sign test. This to spot a difference in mean guilt in "guilt ads" vs mean shame in "guilt ads" for each participant. With the same procedure with shame ads. Specifically a higher guilt score, meaning this should be a one-sided test. 

name   mean_guilt_guilt_ads   mean_shame_guilt_ads   mean_guilt_shame_ads ...
Pete   6.45                   3.34                   2.48
.
.
.

Regarding :

"I have actually tried both paired t-test and Sign test but both showed only 1 out of 8 advertisements have significantly different guilt and shame score."

Are you performing a t-test on all 8 advertisements compared to something? In that case I would recommend the use of a ANOVA instead. This would result in many groups (16). So it would be preferable take a mean of each participants in this case as well, resulting four groups.

I hope I make some sense, otherwise just ask or ignore this answer :)

Improve this answer
$\endgroup$
3
  • $\begingroup$ 1. Yes, that's how my data look like. 2. Maybe my question is not clear. I need to compare the guilt and shame 'value' for every advertisement, so I think I should not collapse the measure of guilt and shame for 4 different guilt ads into one mean guilt and one mean shame. In other words, I would have 8 Hypothesis: 1. Shame ad (a) type one (b) type two (c) type three (d) type four has significantly higher score of shame than guilt. 2. Guilt ad (a) type one (b) type two (c) type three (d) type four has significantly higher score of guilt than shame. $\endgroup$
    – Lyn Smith
    Commented Mar 20, 2020 at 9:31
  • $\begingroup$ If finding out whether, for example, each of the shame ads has a higher score of shame than guilt is difficult, then I just need to find out whether each of the advertisement has a significantly different score of guilt and shame. $\endgroup$
    – Lyn Smith
    Commented Mar 20, 2020 at 9:36
  • $\begingroup$ 'Are you performing a t-test on all 8 advertisements compared to something?' - That is why when I did this, I used SPSS to find the difference between guilt and shame score for Shame ads type one. I just put these 2 measures on the calculation. I do the same for the other 7 ads. Is it wrong? 'Use a paired t-test alternative a "Wilcoxon signed-rank test" rather than a sign test' - The distributions of the guilt and shame score among the participants are skewed, it's not symmetrical. As I understood tit doesn't fulfill the assumption for this test. Am I wrong? $\endgroup$
    – Lyn Smith
    Commented Mar 20, 2020 at 9:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.