4
$\begingroup$

My report is looking at the effect of belief in conspiracy theories and critical thinking abilities, on engagement in politics.

I've basically got to do a factorial ANOVA and my IV's are:

  • Conspiracy beliefs
  • Critical thinking abilities

My DV is:

  • Engagement in politics

The IV's have been converted into high/low groups based on median average etc. I'm already stuck on my first part of wrangling the data. I have to compute z scores. Do I have to compute them for all of my variables? I am using Jamovi.

$\endgroup$
4
  • 3
    $\begingroup$ Welcome to Cross Validated! Why do you have to compute any z-scores? $\endgroup$
    – Dave
    Dec 8, 2023 at 23:58
  • 4
    $\begingroup$ Also, why did you convert your IVs into high/low groups? That is almost always a mistake. And, once you've done that, z scores are going to be kind of silly as there are only two possible z scores? $\endgroup$
    – Peter Flom
    Dec 9, 2023 at 0:37
  • $\begingroup$ Hey so I didn't do that the tutor of the statistics class did that and prepared the data that way for us basically, and they've indicated that we have to compute z scores but I'm honestly baffled as to why now. Do you think I should compute z scores for the results (political engagement mean) only? $\endgroup$ Dec 10, 2023 at 0:05
  • $\begingroup$ Or just don't compute z scores at all? $\endgroup$ Dec 10, 2023 at 0:14

2 Answers 2

10
$\begingroup$

First off, I echo Peter's sentiments where you should just use the raw data and should not dichotomize continuous data in the way you have. This almost always comes at an information loss that affects the tests you run on it (Royston et al., 2006). As he also noted, one can just use regression here, where your two numeric IVs are simply entered as predictors into the regression (with an interaction term) and the engagement in politics variable is entered as the outcome/DV.

This can all be easily achieved within Jamovi. First you enter your variables as main effects (here I use the Parenthood data from the lsj module in Jamovi, though I first convert dan.grump into continuous form).

enter image description here

To build the interaction, you just need to add both predictors together by hitting Ctrl and selecting them together in the Model Building section, then putting them in the same block as shown. You now see the * symbol which denotes an interaction:

enter image description here

The point estimates show that the interaction isn't making much of a difference, but as Peter noted it may be useful to visualize this in some way. One way is to show the simple slopes (see Cohen et al., 2003 for more details), which can be generically visualized with the estimated marginal means (EMMs) section (which estimates the EMMs for categorical variables if they are not continuous like they are here).

enter image description here

The simple slopes show the interaction is very weak. If there was an interaction, the lines from baby.sleep would vary considerably, but here they are almost exactly the same.

References

  • Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed). L. Erlbaum Associates.
  • Royston, P., Altman, D. G., & Sauerbrei, W. (2006). Dichotomizing continuous predictors in multiple regression: A bad idea. Statistics in Medicine, 25, 127–141. https://doi.org/10.1002/sim.2331
$\endgroup$
3
  • $\begingroup$ This is incredibly helpful thank you! $\endgroup$ Dec 10, 2023 at 0:11
  • $\begingroup$ Great! Feel free to accept one of the answers here by hitting the checkmark next to it (if it has answered your question). $\endgroup$ Dec 10, 2023 at 0:13
  • 1
    $\begingroup$ Oh sure sorry! Just getting used to the site thank you $\endgroup$ Dec 10, 2023 at 0:20
7
$\begingroup$

I hadn't heard of Jamovi so I Googled. Seems like it is a way to make R easier to use. (per Wikipedia).

In any case, what I would do here is use the original scores and run a regression, possibly adding an interaction between the two IVs, and maybe looking at splines of those variables, as nonlinearity seems possible.

First, though, I'd make graphs.

$\endgroup$
5
  • 3
    $\begingroup$ "First, though, I'd make graphs." (+1) $\endgroup$
    – Galen
    Dec 9, 2023 at 0:44
  • 3
    $\begingroup$ (+1) Jamovi is more like an SPSS clone written with R that is free and looks a lot better (there is also another version called JASP which is a long story but functions similarly). The code from Jamovi can actually be ported directly into R with the jmv package but not vice versa afaik. I found it to be a very useful program until I eventually moved to R. $\endgroup$ Dec 9, 2023 at 1:18
  • $\begingroup$ I'm not quite sure I would second "make graphs". What for? To select the model based on what we see? That way p-hacking lies. $\endgroup$ Dec 9, 2023 at 13:28
  • 1
    $\begingroup$ I'm not understanding how visualization necessitates p-hacking. $\endgroup$ Dec 9, 2023 at 13:57
  • 1
    $\begingroup$ This is incredibly helpful thank you so much! And it's basically part of my AP statistics class and it looks like they're asking us to compute the z scores which I'm really confused about now you've said that tbh! Thanks so much for providing such a detailed reply, can't tell you how much I appreciate it as I have been a bit baffled tbh! $\endgroup$ Dec 10, 2023 at 0:03

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.