0

I am about to commence data analysis, but am slightly unsure about which test to use.

My experiment concerns persuasion where participants view figures on screen making speeches. Thus must choose which figure won the argument. My design is 2 x 2 (left vs right) x (first vs second). All participants view the same arguments and must choose left vs right (as most persuasive), but in one condition, the right figure goes first and in the second condition the left figure goes first. Participants are either in group 1 or group 2.

Participants view firstly a text on screen - explaining a scenario, they are then presented with figure that give an argument for and against. All arguments are the same across all participants, however condition 1 - figure 1 always goes first - condition 2 - figure 2 goes first. Hypotheses tested are 1) that figures positioned on the left will be more convincing and 2) order may influence decision (i.e., arguments presented first = greater persuasion).

Would a binary logistic regression be the best test here? Step by step would be most helpful.

Thank you

user332893
  • 1
  • 2
    Are participants measured/tested more than once ? Also, are the same figures viewed by more than one participant, or does each participant see unique figures ? – Robert Long Aug 21 '21 at 18:32
  • You don't seem to mention any hypotheses that you want to test. – Glen_b Aug 22 '21 at 03:12
  • Hello both! Many apologies for the lack of detail. Participants view firstly a text on screen - explaining a scenario, they are then presented with figure that give an argument for and against. All arguments are the same across all participants, however condition 1 - figure 1 always goes first - condition 2 - figure 2 goes first. Hypotheses tested are 1) that figures positioned on the left will be more convincing and 2) order may influence decision (i.e., arguments presented first = greater persuasion). – user332893 Aug 22 '21 at 17:24
  • Please add new information as an edit to the post, and not only as a comment. Comments get often unread! – kjetil b halvorsen Aug 22 '21 at 20:04
  • Thank you, halvorsen. :) – user332893 Aug 24 '21 at 15:51
  • How many scenarios do you plan to use? How many subjects do you anticipate? Finally, the scenarios differ in content and the support for one side, eg Left, might vary as a result of the content. This means the probability of selecting L or R is not constant across the scenarios and simple binomial models might not be appropriate. Are you really interested in the sequence? Or in the perceptions of the participants? – David Smith Aug 24 '21 at 16:46
  • Hey David, thanks for your comment. There are 21 scenarios (1 per trial + left/ right argument). Approximately 70 subjects. Indeed, the likelihood of selecting right vs. left, is what I am interested in (e.g., spatial effects on judgement). Additionally, whether order influences this. Thanks again! – user332893 Aug 24 '21 at 17:32

2 Answers2

0

I don't think you need a logistic regression. The outcome is binary (Left/Right is Most Persuasive) if I have understood correctly. The hypothesis is about the order in which arguments are presented.

We can organize the results into a 2x2 table

Left First Left Second
Left Most Persuasive a b
Right Most Persuasive c d

Here, $a$ people who observed left make arguments first thought left was the most persuasive. $c$ people in that same group thought right was most persuasive.

To test if the order in which arguments are presented affects the proportion of people who think left is most persuasive, we can use any number of tests in this answer though I would be preferential to a test of proportions + a Wilson Confidence Interval.

Demetri Pananos
  • 36,121
0

Use a nonparametric method! They're robust and make very few assumptions about your data. They're also weak, but better to err on the side of caution.

For your case, you can run a simple permutation test to obtain a p-value.

user3716267
  • 674
  • Your answer doesn't address OP's hypothesis. What nonparametric method? Permutation test of what? – Demetri Pananos Aug 24 '21 at 16:12
  • Randomly resample the data into a table of the proper shape to approximate the null. Count how many times we observe a difference larger than the experimental results. Another benefit of doing it this way is that it's really obvious what null you're actually testing, because you're constructing it manually. – user3716267 Aug 24 '21 at 16:14
  • No, I understand what a permutation test is and how it works. I'm saying it might be better to perhaps show OP how this test might work for their problem. Maybe you could do a simulation and perform the test for OP. – Demetri Pananos Aug 24 '21 at 16:17