Continuous by Categorical Interaction in Linear Regression (SPSS Tutorial Video #27)

[@rachelgolla2543]

This was so helpful. Commenting for the algorithm so other people can find the help they need.

[@DataDemystified]

Thanks! I’m so glad it was helpful for you!

[@gracedenney3061]

SOOOOOO helpful, thank you!

[@mfarahanynia408]

It is really helpful. I have a question: I have four independent variables (task condition with two levels, task time with two levels, task complexity with two levels, and anxiety scores) with one dependent variables (writing quality). I want to see how anxiety predicts the relationship between the categorical variables (and their interactions like task compleity and task time) and writing quality. What kind of statistical test should I run? Multiple regression (stepwise)? Should I follow your steps for the interactions? Thanks in advance.

[@jingliu6644]

Fantastic prof!!! really helps!!!

[@DataDemystified]

Glad to hear that!

[@bethanycorbett8291]

Did you ever make the video re the follow up procedure for a significant continuous x categorical interaction? Many thanks.

[@charlottedemoor5212]

Thank you for this video, it is very helpful and easy to follow! I've used this video as guidance for how to interpret an interaction with own data and would like to write up my results. Please could you reply to this with an example of how you would write up the results in this video, preferably in APA format? Thanks!

[@DataDemystified]

Hi Charlotte. Thank you for your comment. I'd strongly suggest finding a published paper that reports a 2-way interaction like this one and copying their formatting. Not only will that give you a good sense of what APA style requires, but also give you a good sense of how to think about the key parts of what you are reporting. If you open up, say Journal of Personality and Social Psychology (JPSP), you'll find countless examples of this type of interaction and how it is reported. Good luck to you.

[@jorisjannink6952]

Hello, I've got a question, as you said the topic of having multiple categorical levels which would result in multiple dummy variables would be for another video, I wonder if you already made that video?:) I'm doing a mutiple regression analyses using 1 continous variable and a categorical variable with 4 levels, which I have made dummies for, however I can't find how to do the interaction using mutiple dummies anywhere online, I hope you can help me out, it is for my master thesis :).

[@DataDemystified]

Hi there. Sadly, no, I haven't made that yet. Here's a quick tip: don't make 4 dummies for 4-level variable. You always make N-1 dummy variables (3, in your case). To interact those with the continuous variable, you'd then make 3 more new variables (Continuous x Dummy1, Continuous x Dummy2, Continuous x Dummy3). You'd then include ALL of it (continuous variable, dummies, and interactions) in the regression. The catch, though is that the dummies all are relative to a different level (whichever level you didn't make a dummy variable for), so the interpretation of the interactions is that influence of the continuous variable, on the DV, is different when comparing the referent level to the dummy variable level. I hope that helps! (PS. I probably should make this video sooner rather than later!)

[@jorisjannink6952]

@@DataDemystified thank you so much that really helps!! :-)

[@b0tslayer]

@@DataDemystified in this case you describe, what do you include for "set markers by" to get the plot? Just the regular categorical variable?

[@DataDemystified]

@@b0tslayer That's exactly right!

[@georgiosbaxevanis7811]

@@DataDemystified Hello, thank you for the amazing video!! Have you finally made the new video with multiple categorical levels? Thank you!!

[@FionaPlaysGaming]

Thank you for the video. I have a question about the scatterplot. If we use non-centered variable for the predictor, will the scatterplot change? I am not quite sure whether we should use non-centered variable or centered variable when we report it in the paper. Non-centered variable makes it hard to interpret because you cannot see the previous values now.

[@reutbinyaminnetser386]

Hi, thanks for the video. Can you please show how to report these results?

[@arambulm3]

Hello Dr., I have a problem with some data analysis, let me explain: a bioassay was conducted to determine which diet (treatment) is more effective in shrimp growth. A quadratic regression was performed (SPSS v26), but now we want to perform a broken-line regression to see what the breakpoint is between diet 3 and diet 4. Could you help guide me on how to do this in SPSS?

[@TheSGPhoenix]

awesome video, thank you very much! So does that mean that when you have significance in the ANOVA table, you have an interaction? Or does the significance for the coefficient also has to be < .05 to be valuable?

[@DataDemystified]

Glad you like the video! The ANOVA in a regression just tells you if the model, overall, has any predictive power. The interaction is determined by the specific interaction coefficient.

[@TheSGPhoenix]

@@DataDemystified thanks for the answer! This means that if my interaction coefficient does not have significance, I don't have an interaction? The slopes of the plots don't really look different, but this interaction is new to me.. So I find it hard to interpret my results

[@DataDemystified]

Correct

[@Evan-cn1ki]

Really thanks a lot!

[@cesarhoyosalvarez6569]

Thank you for this excellent video. Quick question, is it possible to add other continuous variables to the model?

[@DataDemystified]

Thank you! Yes, absolutely. You can make your model as complex as appropriate to your research question. For instance, if you have three predictor variables (IVs), your model could be: Y = A + B + C + A*B + A*C + B*C + A*B*C. What's critical is that if you include a higher order interaction (e.g. A*B*C) you MUCH include all the subordinate interactions (e.g. A*B + A*C + B*C) in the model. Not doing so will make the Beta coefficient for the three-way interaction uninterpretable.

[@cesarhoyosalvarez6569]

@@DataDemystified Gotcha. Thank you. Can my model still be valid even if I choose not to include a higher-order interaction and I just input the subordinate interactions?

[@DataDemystified]

@@cesarhoyosalvarez6569 Absolutely!

[@cesarhoyosalvarez6569]

@@DataDemystified Got it. Thank you so much. Oh and one last thing, if I did a previous Factor analysis do I need to center my variables/ I also standardized them 0-1. None of them are correlated

[@DataDemystified]

@@cesarhoyosalvarez6569 Centering is generally a good idea in regression analysis (though not always needed).

[@ranjanachannoo7280]

What would be your hypotheses?

[@rubenmultiking]

Firstly, Thanks a lot for this video series, im doing my masters thesis and i have had some offtime from academics so this really helps! I have one question, does anyone know a source to explain the importance of centering more extensivly? I would like to know more about why it is imortant!

[@katarzynagustavsson4983]

What if I have multiple groups turned into dummy variables but I also have multiple continuous predictors? What should the interaction variable be then?

[@DataDemystified]

You would have a lot of interaction terms. EVery single dummy coded variable would be interacted with every single continuous variable (assuming you want to test all possible interactions).

[@katarzynagustavsson4983]

@@DataDemystified And separate variables need to have significant coefficients or do I only look at the entire model when writing down the regression coefficients as the equation in the video?

[@katarzynagustavsson4983]

@@DataDemystified I would greatly appreciate a response to my comment in this thread.

[@nassibbb2]

thats great, thanks!

[@minaehab7626]

Thank you so much for your videos . But I have Q on this video , what if we inversed the codes for both M,F (M:0 , F:1) i think it will get different result according to the equation (as per my understanding) @datademystified