# Insignificant two-way MANOVA results - what do I do next?

I am very very new to stats. I decided to try and do a MANOVA in SPSS, my between-subjects IV is gender and my within-subjects IV is the two conditions in which all participants took part. My DV is the perspective (see below) from each of the two conditions.

My MANOVA showed that p>0.56 for all of the multivariate tests for Y1*Y2 (gender*condition) so if I'm correct then this indicates that there isn't a statistically significant interaction between gender and the conditions on scores?

What do I do now that I have found this - is it worth reporting if it is not statistically significant? I don't understand whether this means that gender doesn't seem to have any influence on the scores received in the two conditions, or if there is another test that I can do to try and find a statistically significant interaction. I would like to report my results from my MANOVA still but then perhaps carry out another test but I don't know what other tests I can do or whether I should just leave it.

I'm also confused as there are differences in the scores received in the two conditions by each gender but then I think the MANOVA contradicts this?

My apologies, this is my first time ever really trying something stats-related and as it is clear, I am struggling a lot. Thank you for your time.

EDIT: The study was looking at the perspective participants gave when describing a photo, whether it was from their own perspective looking at the photo or from the perspective of the person in the photo. The two conditions were the same photo, one with a person in it and one without to see if people would be more likely to give the egocentric (self) perspective when there was not a person in the photo. In the study, I wanted to look at gender differences in perspective-taking. There were 169 participants in this repeated measures study - 21 males and 148 females. The perspectives were coded as numbers (e.g. if they described the image from self-perspective, it was coded 1 and if they didn't then it was coded 0. All of the variables were nominal. I carried out a MANOVA test, I would have chosen the multivariate test that is more robust when it comes to violations as the Shapiro-Wilk test indicated my data wasn't normally distributed (p<0.001). The skewness was -.91 (moderately skewed) and kurtosis was -1.17 (platykurtic). However, for all of the multivariate tests, the significance was p>0.54 which clearly demonstrates no statistical significance. I don't understand whether or not this is a good thing nor what no statistical significance indicates. I have tried searching it but that just confused me more.

Thank you so much!

• If your outcome variable is binary (0/1) then simple linear models like MANOVA are inappropriate as they assume continuous outcomes. It seems that what you need instead is a logistic regression model, with the binary outcome (perspective chosen) regressed against gender, photo type, and their interaction.
– EdM
Dec 8, 2019 at 19:14
• Welcome to the site. I'm not sure how to say this but, if you are really very, very, very new to stats and this is the first time you are doing something stats related then I suggest you find someone with experience to help with this. MANOVA isn't the place to start with statistics. Dec 9, 2019 at 12:58

The perspectives were coded as numbers (e.g. if they described the image from self-perspective, it was coded 1 and if they didn't then it was coded 0).

MANOVA is for continuous outcomes and is not an appropriate test for evaluating binary outcomes like yours. So you should not pay much attention to the results you got with that approach.

Some type of logistic regression model would be better, essentially a model of the probability* of describing from self-perspective as a function of gender and the presence of a person in the picture. As your hypothesis is that gender will influence how the presence of a person in the picture will affect the probability of describing the picture from self-perspective, you would include in the model an interaction between gender and presence of a person in the picture as a test of that hypothesis. As you have multiple measures on the same individuals, you could take that into account with a mixed model allowing for differences in probabilities in choosing the self-perspective without a person in the picture.

If this is the first time that you are trying to do serious statistical analysis, however, you should follow the advice that Peter Flom gave in a comment and work with someone who has statistical expertise. The issues with interactions in logistic regression models and with repeated measures can be tricky, and there might be details of your experimental design that could suggest alternate analysis approaches. Going forward, it's always best to have such statistical consultation before undertaking the experiment, to make sure the design can provide clean answers to the questions you are asking.

*more precisely, the log-odds

I'm assuming the design and analysis is all well justified...

is it worth reporting if it is not statistically significant?

Yes

I don't understand whether this means that gender doesn't seem to have any influence on the scores received in the two conditions

In short and without knowing more about the study, it means that your results are completely consistent with random chance.

or if there is another test that I can do to try and find a statistically significant interaction.

Statistical significance is not something you go looking for. Assuming your study was well designed and the MANVOA was the appropriate analysis to perform, then you've done all you can do. Purposefully performing additional tests in order to get statistical significance in some respect would be p-hacking.

Could you post some of your data and tell us a little more detail about your study?

• Thank you! I will add some more detail now Dec 8, 2019 at 18:33