SPSS - Double Moderation with PROCESS and Covariates (Model 2)
ฝัง
- เผยแพร่เมื่อ 12 ก.ย. 2024
- Lecturer: Dr. Erin M. Buchanan
Missouri State University
Summer 2018
You will learn how to use the new version of the PROCESS version 3 plug in for SPSS by A Hayes. In this video, you will learn how to run a two (double) two-way moderation model with covariates, as well as data screening and power for an analysis with covariates. This model specifically is model 2 with two moderators and their simple slopes.
Tomorrow the R version of this same analysis will be posted - so check back in for that!
Get the materials and other information at OSF: osf.io/ns6jz/
Hello. You are such a lifesaver. I've been trying to figure this out. The two videos on moderation really helped me improve the results section of my dissertation.
Thanks and glad it helped!
Excellent videos! Is there a way to create this graph in Excel? I find the SPSS-produced charts a little tricky to interpret and customise?
Agree! I believe you could export the numbers into excel and create a grouped bar chart as well.
This is extremely helpful, really. I do have a question, in the newest version of SPSS and PROCESS, there are two option for mean centering (all variables that define products and only continuous variables that define products). what is the difference between them and which do you recommend using?
I would think only continuous variables! Seems weird to do categorical variables?
Hi Dr. Buchanan,
Thank you so much for your awesome presentation. My colleagues and I are wrestling trying to understand the practical differences between model 2 and model 3. I know model 3 is a three-way interaction and model two is just two interactions. But the way PROCESS sets forth the data under the "Conditional effects of the focal predictor..." it looks similar to a 3-way interaction. I am sorry is this is a simplistic question but how would you explain the difference?
Model 2 is just TWO two-way interactions (m*x and m*w), while model 3 includes the full three-way interaction (m*x*w).
@@StatisticsofDOOM hello Dr. Buchanan, your videos have helped me with my researches so much. For this double moderation analysis, it makes me wonder what the difference if we analyze two moderators using model 2 and analyze it using model 1 but two times with different moderator?
@@awkayffafin Those are definitely different analyses, as the first includes all variables in the model, while the second does not. The different variables in each model will produce different results.
Hi Dr. Buchanan. Thank you so much for a very helpful presentation. I have a simplistic question for which I apologize but I thought I should try to clarify. When PROCESS gives the "Conditional effects of the focal predictor at values of the moderator" you said that PROCESS was not breaking the results into categories but looking at the entire database-you added something like we are just pretending to look at the low, mid and high values of the moderators. What I want to say is that when Moderator 1 is low and Moderator 2 is low--then the association between X and Y is ____(e.g., positive and significant). But when Moderator 1 is high and Moderator 2 is low then the association between X and Y is...whatever. Would that be correct?
Answering the right question now: yes that would be correct - I often provide people examples what i mean by low/average/high ... so low is 1+SD below the mean in this example, so you could provide that number to people to help conceptualize the idea.
Hey dear thank you very much for your helpful video, I have a quick question, I have 2 moderation M1 and M2, each one has two conditions, only one of the moderations, M1, is statistically significant, so how should I interpret Conditional effects of the focal predictor..." ? I mean should I mention the impact of the non-significant moderator , M2, or I should not talk about that at all? there is also another problem, from the Conditional Effect part in the results, it is clear that only one of the two conditions of M1 is significant, so should I talk only about the significant one?
Thank you in advance for your time.
I would mention all effects for transparency. Just make a table of all coefficients, t, p values.
Hi, may I please seek your opinions, if my model is Model 8, can I use the same method shown in the video to determine my sample size (i.e., effect size = 0.07, significance level = 0.05, power requirement = 0.80)? Thank you so much!
Basically, yes, but you'd want to change to the appropriate number of predictors for Model 8.
Thank you for the video. I have 3 moderators in my model. Can you suggest which Model to use from Hayes template?
I’m not sure which one can handle 3 moderators. I don’t think the current process can do more than 2.
Thanks for the video! Would like to clarify on the outcome for the 'constant coeff' in Model 2. Does this mean that the value represents the mean score for both the moderators, W and Z? Thanks.
That should be the intercept if I remember correctly, which is the average of Y if all predictors are zero.
16:40 PROCESS regression /w covariate
It was a very helpful video, thank you very much. Are there any source about how we can create a table in Apa style?
I mean other than the APA manual? I’m not totally sure what you are asking.
Are you able to a double moderation with a covariate in PROCESS, if your covariate is categorical (i.e., gender)? Thanks!
You should be able to! Just be sure to mark that it's categorical.
@@StatisticsofDOOM Thank you so much that is really helpful, I have one more question. Is it possible to do a double moderation with one moderator being categorical and one being continuous in PROCESS?
@@TaylorD-l9v I think so! You just have to mark the categorical one.
HI! thanks so much for this video, it was really helpful. I have a quick question though... I realise at around 29:28 minutes, it is "At low M1, increasing M2 leads to X-> Y". However, I want to study the effects of "At low M1, how does it affect X-> Y" ;and "At high M1, how does it affect X-> Y"; followed by "At low M2...". In this case, is it possible if I utilised Model 1 of PROCESS, but to carry it out on two occasions?
Sounds like it if you don't want the M1 and M2 to interact. (or rather to be in the same model)
I've recently conducted a Model 2 moderation with PROCESS v3.4. In this version, there is a row for interpreting the interactions of BOTH (presumably X*W and X*Z taken together). Any guidance on how to understand this "BOTH" row? Is it simply the moderating effect of a composite of W and Z?
That's what I would guess as well or might it be the three way interaction?
I've sorted it out and it is the combined effect of the two moderators, which isn't all that informative without examining the interaction or looking at the individual effects of each. Thank you for your videos! These are a lifesaver.
excuse me, could you share some details about the BOTH, i've conducted a Model 2 moderation with PROCESS 3.5. but only one of the int(X*W ) is significant, the other int(X*Z) is not. But the BOTH is significant...you said that" isn't all that informative", What does it mean..
Or the Both significant is just because one of the moderator is significant?....
Thank you so much! May I ask another question?
What are the statistical assumptions/requirements for a double moderation that must be met? Maybe I need a break, but I could not find anything on this topic (even after searching for 1h :D ). I am even thankful for paper or book recommendations!
The video goes through these assumptions at the beginning - it's the same as regression: linearity, homoscedasticity, normality, additivity.
@@StatisticsofDOOM uff, thanks a lot! I thought there were more assumptions than a "normal" regression. :) But this solves the problem that I could not find any other assumptions on the internet or in a textbook that I found :D
Thank you for the amazing video! I do have one question concerning a model I am running myself. In my study X and the two moderators all have significant main effects. Also, the Interaction of X and M1 is significant, but the interaction of X and M2 is not significant. Can I interpret the results in the same way as above? (a single moderation for X and M1 shows a non-significant interaction)
I would just interpret the single interaction and not the full three way interaction.
Hello and thank you very much for the amazing and very helpful video! I just have a small question, which did not get covered here and to which I could not find the answer somewhere else. In my project I could prove that X leads to Y. When I add the two moderators, they are both not significant (the interaction effects are also not significant) but the causal relationship between X and Y becomes not significant as well. Does this mean something?
That implies intercorrelation between X and Ms, so that when you include M, the shared variance is large so X is no longer significant. And possibly that the relationship between X and Y is small, so it flips when you include M.
Hi, May I ask one other question. Is there any reason why someone can't use the PROCESS default conditioning values of 16th, 50th and 84th percentile?
You could - just depends on your preference (I was taught simple slopes, but those percentiles could be useful too).
@@StatisticsofDOOM Thanks so much! Just one last clarification; doesn't using the PROCESS default percentages still produces simply slopes -just at different percentages? Sorry for the fundamental questions...
@@saramason7876 It's still simple slopes yes, just a specific percentile points rather than the traditional +1/-1SDs. If you use those, I would definitely make it clear you are using percentage cut points, as I think the term simple slopes is often taken to mean the SD approach.
Hi, Thank you for all the video tutorial, I have one relevant question to moderated mediation. Using PROCESS macro, can, and how can we calculate the change or increase in indirect effect of X on Y with 1SD increase in X at the high value of moderator via mediator. Other words, can we measure %age change in Y with the 1SD increase in X via mediator at high level of moderator.
That would be a different model, as this one is only for interactions. You would want to pick the right model version, and then it will give you the the simple slopes and the mediation at those simple slopes.
Thank you so much for this video! It has been incredibly useful to me!
Do you have any examples of how you write this up in an APA style report?
Not specifically this analysis but stuff like it can be found here: osf.io/vf5gh/
Hello, if one interaction was significant and the other one is not, then can we interpret the conditional effect box or not?
thank you so much , this video is very helpful
I would only interpret the conditional effect for the significant interaction.
Thank you for the clear explanation. However, my study is different in some way and I want to know whether something can be tested with model 2 or if I should separate the data and use model 1 twice. My X (0 = unhealthy version, 1 = healthy version) and M1 (pizza vs sandwich) separated my data in 4 groups (unhealthy/healthy sandwich/pizza). However, M2 is health consciousness and I want to test for both product types (pizza vs sandwich) whether health consciousness has a moderating effect on the X->Y(perceived tastiness) relationship. I know I can see the conditional effect of X on Y for different values of M1, but can I see the interaction effects of X*M2 for different values of M1? Or should I split the data on M1 and test this with model 1 twice? Thank you!
Try model 4 which allows for two-two way interactions.
@@StatisticsofDOOM isn’t model 4 mediation?
@@nickthape1479 Yes apologies, I mean model 3. If you want the three way effect, you'd have to do it by hand. Then I would recommend ANCOVA, as it's the same idea.
@@StatisticsofDOOM thank you, this really helps me!
Hi, Thank you for the explanation. I have a small clarification required. In the model, when you mentioned X (Q151) the Independent variable does not predict the Y (Q11) dependent variable because it is insignificant. Does that mean, there is no significance due to the moderator effect but still there is a relationship between X and Y.
Yes, this can happen when there is not a main effect but is an interaction.
Is this considered a moderated multiple linear regression?
Yes, with two moderators.
Very helpful video, thanks! I have a question; can you also do the above with categorical (2 categories for each variable) iv & moderators?
Is there a reason not to use ANOVA in that scenario if all the IVs are categorical?
A very helpful video, Maa'am. Could you share a research article using Model 2 of PROCESS Macro?
I don't know that I have a specific one - I think there might be some on Hayes' website.
Hello thank you so much for your helpful video! I have a quick question: I have run model 2 in process version 3.5.2 and selected the "Johnson-Neyman output" option. However it will not provide any output that seems to be related the the J-N technique. Will the J-N technique not run for model 2? I saw in your other video "SPSS - Moderation Analyses with Simple Slopes + Process" that when you went to select the J-N option (in an older version of process) the option said "Johnson-Neyman (Models 1 and 3 only)". I appreciate any clarity you can provide!
To clarify, in the model 2 that I ran, Interaction 1 is significant and interaction 2 is not significant (if this info is helpful!)
@@user-fo8kg4ux6n I'm unsure honestly, but I know there's an option that you have to tell it to probe the interaction at a specific p value. Maybe try upping that number? I don't know if it'll run for Model 2 either.
If you have FOUR dummy-coded covariates, for ethnicity, and only one of the ethnic groups is significant (Biracial), how do you interpret? The moderator is significant (social support moderating stress on depressive symptoms). Thanks!
th-cam.com/video/Zv19sslm-S4/w-d-xo.html - maybe try watching the first half of this video (not the R part, but the dummy coding explanation).
@@StatisticsofDOOM thank you! Watched. I get the dummy coding piece, but I'm not sure how to interpret PROCESS model 2 with a significant covariate. Any advice would be great. THANKS!
@@sallyatheran1133 Great! I would just say the covariate is significant - if it's the ethnicity variable it means that there's a difference in the means for the DV between the two ethnicity groups. That difference is adjusted out in the other variables.
@@StatisticsofDOOM THANK YOU so much! You are so helpful and a gem.
Greetings!
Are there studies where the main effect (X), does not predict (Y) and when Moderators(M and W) interact with X, produces a significant interaction? Or studies similar to your example?
Thank you, this video is really helpful.
Oh I'm sure there are ... that would imply that X only has an effect on Y when considering certain levels/areas/zones of M or W. Not sure if I have an example of that, but it would basically be like at low M, X does predict Y (negative), at average M, X does not predict Y, but maybe at high M, X does predict Y (positive). Overall, X doesn't predict Y because it all washes or averages out.
@@StatisticsofDOOM Thank you very much! It is a big help!
Your videos are awesome! More power to your channel! Best of luck!
Is it possible to run this with 3 moderator variables? (A triple moderation, if you will). And if yes, what would the advantage of that be over running three separate model 1 moderations?
I don't believe process allows for more than 2 moderators (which is already 3 way interactions). Doing it as the interaction allows you to see all the possible combinations, while only doing two way interactions one at a time only shows you each of those two way combinations individually. I don't think that three and four way interactions are always understandable, but it depends on the research question.
For my analysis, I also use Model 2, but my Model Summary looks completely different. For example, instead of having R and R^2 my model summary gives -2LL and ModelLL. Do you have any idea why that is?
The versions of process may be different, that looks like loglikelihood rather than R2.
@@StatisticsofDOOM do you also know how I can change it to R2
@@kakkie40 Do it show you residuals/deviation? You can also calculate it from the F statistic - our app does it: www.aggieerin.com/shiny/mote/
I wonder what if the p value in model summary is significant, the "model section" and in "test of highest order unconditional interaction" are bot nonsignificant, but the output shown the "conditional effect of focal predictor at values of the moderators" and p value in here is significant? Can I say the factor has moderating effect on the relation?
To say if there is moderation: you would expect the p value for the interaction term to be significant.
You could argue that the overall model doesn't have to be significant, but you can't argue for an interaction when the interaction term is not significant. The conditional effects may be significant, but not different from each other (which is the key component to claiming an interaction).
For example conditional effect1 may be .45 which is greater than zero, but not different than conditional effect2 which is .40 ... that implies only a main effect of the variable.
@@StatisticsofDOOM I see, thank you.
@@StatisticsofDOOM sorry, can I ask you one more question? the age split to 3 age groups and use as multi-categorical moderator, so the younger age group become the reference for the middle age group (W1) and older age group (W2). If the interaction_1 (X*W1) is significant, but the interaction_2 (W2) and the X*W are both nonsigifnicant. is that means age is not the moderator but middle age group itself is the moderator of this relationship? or should I say, x has a stronger effect on Y for the middle age group when compared with the younger age group
@@Cjstistajh sounds like it only moderates for the young to middle comparison.
@@StatisticsofDOOM thank you very much.