Adding APA-formatted stats to results summary .pptx

[hagenaue]

I compiled a summary of the effects of bred line and treatment on each of the basic (non time-series) variables in a powerpoint for Angela. It can be found using this MBox link: https://umich.box.com/s/sn70kx25rhznlw90fs1moiewee5owygv

Currently the results summary only contains p-values, but to write up the results in paper format we will have to use APA formatting, which looks like this: F(degrees of freedom for variable, degrees of freedom for residuals)=Fstat, p=p-value e.g., Main effect of Line: F(1,128)=188.70, p=6.667e-05

The F-statistics and degrees of freedom for the permutation-based ANOVA (w/generation as a co-variate) results for each variable can be found in this MBox folder: https://umich.box.com/s/a7bsvapvz2yaajqnurvh1gkpu5rglxsu

[hagenaue]

I've now added a summary of the effects within the social defeat time-series variables to the powerpoint as well (at the end). As before, I've only quickly summarized the significant effects and provided p-values, but the final publication will need F statistics and degrees of freedom (discussed above) . The F-statistics and degrees of freedom to accompany the p-values for the time-series variables (from an ANOVA summary of a multi-level regression model centered on Day 4 of social defeat with generation as a co-variate) can be found here: https://umich.box.com/s/aogdbrvoe5nrzfg4z88mkm1hef3lplql

Update: Hold off on this while I double-check why the stats are listed as Chi-Square...

[hagenaue]

Alright, I found more information about the output from the anova() summary of the multi-level regression models for the time series variables. Apparently the output is from a likelihood ratio test ("LR") instead of an F-test. https://www.rdocumentation.org/packages/car/versions/1.0-9/topics/Anova

This should be summarized similar to Chi-Square statistics by reporting the df associated with the variable, the full sample size, and the Chi-square statistic (which is abbreviated with a greek X that is squared (superscript 2): X^2 (df1, N=sample size)=Chi-Square Statistics, p=p-value. Example: Χ^2(1, N = 75) = 0.89, p = .25. https://www.statisticshowto.com/reporting-statistics-apa-style/

[evrich17]

I started going through these this morning to add all the F statistics but there are a lot of p values listed on the powerpoint that are not matching up to the value that is listed in the folder on MBox, so I'm not really sure if these F statistic values are accurate or if there's something I'm missing...?

[hagenaue]

I started going through these this morning to add all the F statistics but there are a lot of p values listed on the powerpoint that are not matching up to the value that is listed in the folder on MBox, so I'm not really sure if these F statistic values are accurate or if there's something I'm missing...?

Can you provide an example?

[evrich17]

for example, for the SITest_SocialInteraction_Ethovision, the main effect of enrichment p-value is listed as 0.06207 on the powerpoint, while it is listed as 0.0002000 on MBox. There are several others like this and in some places it looks like the generation co-variate value may have been switched with the interaction of Line x Enrichment value (e.g., slides 3 and 4). I wrote in most of the F statistics that I think are correct but did not change any of the p-values, and I highlighted all the values that appeared to have inconsistencies in red on the powerpoint

[hagenaue]

Awesome - I'll take a look and see if I can figure out what happened. It's possible that they are just typos. It is also possible that some of them may differ a little because permutation-based analyses produce slightly different results each time they are run, and I re-outputted a simplified version of the results for Mbox.

[hagenaue]

Alright, I found out more about the "N" that is supposed to be reported with the Chi-square statistic: It is just the sample size for the experiment, which for the time-series variables should probably be reported as # of subjects.

....Which reminds me that the sample size differs for many of our variables in general, so I added some code to loop over all variables in the dataset and output N for each of them, because we will need to report that information somewhere in the paper. The output (Ns for each variable) can be found here: https://umich.box.com/s/4b37yq8rsufelg3cnd0mgwfnquf9ihzd

(relevant reference: https://psychology.stackexchange.com/questions/10482/how-should-i-report-results-of-a-likelihood-ratio-test)

[hagenaue]

I ended up fixing the last few things when I added the stats into a draft of the results section: https://umich.box.com/s/h35ouw2cogfphe3n5zd0wdrqp336o82f