Thank you so much for this tutorial, it's been a huge help while I write my thesis! I do have one question regarding the bootstrap results and I haven't been able to find an answer in any textbooks, papers, or forums so far. I tested 3 hypotheses with the macro (Model 1) simple moderation analysis including H1: regarding the association between the IV and the outcome variable (Y), H2: regarding the association between the moderator (W) and the outcome variable (Y), and H3: whether W moderates the relationship between IV and Y. I encountered some assumption violations (heteroscedasticity, normality and independence of residuals) and therefore decided to use heteroscedasticity consistent standard errors (HC4) and bootstrapping with 5,000 samples. When reporting the results for H1 and H2, would you be willing to provide some advice on whether I should only report the bootstrapped results (i.e. bootcoeff, bootSE, [boot LL, UL]) or if I should also report the t- and p-values from the asymptotic (non-bootstrapped) output as well?
I think both is possible. It depends how you write it down. You could write that you used bootstrapping as an additional safeguard, then you would write both (conventional values and bootstrap). Or you could write that you used bootstrap as a robust procedure and only report bootstrap. I chose the first approach for my bachelor's thesis and, I think, for my master's thesis, too. But please be careful, a violation of the independence of the residuals is to my knowledge not remedied by using bootstrapping.
Thank you very much for the video. I am working on a study in which participants are randomized to three conditions: 1) 1-option condition, 2) 4-option condition, 3) 16-option condition (so this is the IV, with *3 conditions* instead of 2 conditions). The study has an individual difference moderator which is continuous. One DV is continuous and another DV is binary, yes or no. I would like to ask, if it is possible to run mean, +1SD, -1SD analysis (with p-values and effect sizes) and JN method, with Hayes PROCESS in R? If yes, how? Any resource/tutorial/video on that?
I haven't tried this specific configuration yet, but since the moderator is continous it should be possible. For the multicategorical IV you need the additional parameter mcx = 1 in order to tell PROCESS to construct two dummy variables for the three conditions of the IV.
@@RegorzStatistik Thank you very much, that's very helpful! Yes I just tried and this seems to be working. One issue is that the outputs only compare between 1-option condition and 16-option and 1-option condition and 4-option condition, but *not* 4-option condition with 16-option condition. How to set the reference group/condition properly in process? Also, the "effect" at different values of the moderator seem to be the *mean difference* between conditions, what can I do to change that to be more interpretable effect size like cohen d or standardized regression coefficient?
@@siukityeung6583 1. You could change the coding for your IV so that another group is the reference category (you will always get only two comparisons with three groups) 2. For a multicategorical IV you are looking at mean differences. I don't know any way to request d from PROCESS, unfortunately.
@@RegorzStatistik Thank you very much! I changed the reference group and that works. I wonder if one of the DVs is binary (yes or no) whereas the IV is categorical (3 conditions) and the moderator is continuous, is it possible to run *moderated* logistic regression with Model 1 in process, with simple slope analysis/mean +1SD -1SD analyses? Thanks
I guess your interaction wasn't significant? I think the default p-value for printing conditional effects is .10 - for p-values of the interaction higher than that the conditional effects are not printed (and they are not really relevant in the absence of a moderation).
This is a really useful tutorial-many thanks. When I run the code you suggest, I do not get a section in my output called 'Conditional Effects of the Focal Predictor at Values of the Moderator/s'. Is this because my moderation (i.e., the extent to which the interaction between the predictor variable and the moderator variable predicts the outcome) was non-significant?
Whether you get the "conditional effects" (=simple slopes) depends on the p-value of the interaction term. If I remember correctly, the default value is .10, i.e. only for a p-value below .10 you get the simple slopes. This default value can be changed by setting the parameter *intprobe* to a value different than .10.
There is a syntax option for multicategorical variables, c.f. www.regorz-statistik.de/en/process_3_syntax.html#general You'd just have to use the format for the R-syntax, that is: mcw =1 (instead of /mcw = 1 for SPSS) (However, I haven't tried this with PROCESS for R yet.)
I suppose so, but I haven't tried it yet. So I would recommend testing it (with your real data or, if you haven't got that yet, then with simulated data).
Thank you so much for posting this, really helped with my Master's thesis :D
I found this vid helpful! Thanks for making it!
Thank you so much for this tutorial, it's been a huge help while I write my thesis! I do have one question regarding the bootstrap results and I haven't been able to find an answer in any textbooks, papers, or forums so far. I tested 3 hypotheses with the macro (Model 1) simple moderation analysis including H1: regarding the association between the IV and the outcome variable (Y), H2: regarding the association between the moderator (W) and the outcome variable (Y), and H3: whether W moderates the relationship between IV and Y. I encountered some assumption violations (heteroscedasticity, normality and independence of residuals) and therefore decided to use heteroscedasticity consistent standard errors (HC4) and bootstrapping with 5,000 samples. When reporting the results for H1 and H2, would you be willing to provide some advice on whether I should only report the bootstrapped results (i.e. bootcoeff, bootSE, [boot LL, UL]) or if I should also report the t- and p-values from the asymptotic (non-bootstrapped) output as well?
I think both is possible. It depends how you write it down. You could write that you used bootstrapping as an additional safeguard, then you would write both (conventional values and bootstrap). Or you could write that you used bootstrap as a robust procedure and only report bootstrap. I chose the first approach for my bachelor's thesis and, I think, for my master's thesis, too.
But please be careful, a violation of the independence of the residuals is to my knowledge not remedied by using bootstrapping.
Thank you very much for the video. I am working on a study in which participants are randomized to three conditions: 1) 1-option condition, 2) 4-option condition, 3) 16-option condition (so this is the IV, with *3 conditions* instead of 2 conditions). The study has an individual difference moderator which is continuous. One DV is continuous and another DV is binary, yes or no. I would like to ask, if it is possible to run mean, +1SD, -1SD analysis (with p-values and effect sizes) and JN method, with Hayes PROCESS in R? If yes, how? Any resource/tutorial/video on that?
I haven't tried this specific configuration yet, but since the moderator is continous it should be possible.
For the multicategorical IV you need the additional parameter mcx = 1 in order to tell PROCESS to construct two dummy variables for the three conditions of the IV.
@@RegorzStatistik Thank you very much, that's very helpful! Yes I just tried and this seems to be working. One issue is that the outputs only compare between 1-option condition and 16-option and 1-option condition and 4-option condition, but *not* 4-option condition with 16-option condition. How to set the reference group/condition properly in process? Also, the "effect" at different values of the moderator seem to be the *mean difference* between conditions, what can I do to change that to be more interpretable effect size like cohen d or standardized regression coefficient?
@@siukityeung6583
1. You could change the coding for your IV so that another group is the reference category (you will always get only two comparisons with three groups)
2. For a multicategorical IV you are looking at mean differences. I don't know any way to request d from PROCESS, unfortunately.
@@RegorzStatistik Thank you very much! I changed the reference group and that works. I wonder if one of the DVs is binary (yes or no) whereas the IV is categorical (3 conditions) and the moderator is continuous, is it possible to run *moderated* logistic regression with Model 1 in process, with simple slope analysis/mean +1SD -1SD analyses? Thanks
@@siukityeung6583 I don't know, I haven't tried this scenario, yet.
Hello, thank you for this video. I am having issues finding the conditional effects. They do not show up after I run the model.
I guess your interaction wasn't significant? I think the default p-value for printing conditional effects is .10 - for p-values of the interaction higher than that the conditional effects are not printed (and they are not really relevant in the absence of a moderation).
This is a really useful tutorial-many thanks. When I run the code you suggest, I do not get a section in my output called 'Conditional Effects of the Focal Predictor at Values of the Moderator/s'. Is this because my moderation (i.e., the extent to which the interaction between the predictor variable and the moderator variable predicts the outcome) was non-significant?
Whether you get the "conditional effects" (=simple slopes) depends on the p-value of the interaction term. If I remember correctly, the default value is .10, i.e. only for a p-value below .10 you get the simple slopes. This default value can be changed by setting the parameter *intprobe* to a value different than .10.
Thank you for this tutorial. What about if my moderator is a categorical variable with 4 different groups?
There is a syntax option for multicategorical variables,
c.f. www.regorz-statistik.de/en/process_3_syntax.html#general
You'd just have to use the format for the R-syntax, that is:
mcw =1 (instead of /mcw = 1 for SPSS)
(However, I haven't tried this with PROCESS for R yet.)
@@RegorzStatistik ok...thank you! Can I use this moderation with PROCESS if my IV is a dummy coded in addition to my moderator that is dummy coded?
I suppose so, but I haven't tried it yet. So I would recommend testing it (with your real data or, if you haven't got that yet, then with simulated data).