There are very few videos on RUclips that are just as useful. You are awesome! I just wondered. If you want to compare also average and high groups, do you need one more analysis that covers just these groups?
In my case I have a moderator with four levels (four different experimental conditions) that are independent to each other, but the program gives to me W1, W2,W3, instead of W1, W2,W3,W4, why? thank you!
Dear Dr. Erin, Thank you for your video. Given that referece group is a low group, I think the results do not indicate whether the slope of average group (m2) is significantly different from the slope of high group (m3). How can we compare the slopes of m2 and m3? Do we have to reanaylze after chaging the reference group?
Hello there, thanks for the helpful video! I have a problem regarding the output SPSS gives me. For some reason that I cannot identify I do not get the output on the "conditional effects of the focal predictor at values of the moderator". Instead the output just stops at the "test for highest order unconditional interaction". Have you ever faced a similar issue or suspect a reason? Thanks!
That implies that the p value for the interaction is not significant. There's an option to show those effects at a higher p value, which you need to change to if you want to see it when non-significant.
Dear Erin, thank you so much for your videos - they are super helpful! A quick question please, what if my X is categorical with 2 levels high vs. low, my M is 2 levels as well with high vs. low and my Y is two different types of products (hedonic vs. utilitarian), would I still be able to use simple moderation 1 and dummy code all my 3 variables, or should I use ANOVA, or any other analysis? Thanks so much for your response in advance!
I think it's helpful, but still figuring out how exactly to do it... I have a moderator of 4 different categories (without order). How would I do that? Same as you and code the last one as 1 & 1?
The new version of process has categorical moderators I believe, you just have to indicate it's categorical so you get simple slopes by group, rather than low, average, and high.
Dear Dr.Erin, thank you so much for your video！Do I need to encode(e.g., Indicator) both the independent variable and the moderator variable when both are categorical variables with three groups, and the dependent variable is continuous? If so, what specific metrics(e.g., X1*W1) should I look at in the subsequent analysis?
Thanks a lot for your videos! I am stuck with a model where I have a categorical IV (more than two levels), one continuous moderator, and a continuous DV. Can I use PROCESS to run such an analysis? If I use ANOVA with a mean-centered moderator, I think I would only be able to talk about the effect at the average value of the moderator, unlike PROCESS where I am able to talk about the effect 1SD above and below the average score of the moderator (which is a pretty nice thing I have learned from you). What would be your suggestion? Hope you have a great 2022!
If you are using the newer version of process, I believe so - I'm not in this video but you can use categorical M in many new places, so try it and see what happens!
Hi. Thanks so much for this video. I noticed that your categorical variable has 3 levels. Is there a problem using a categorical variable that has only two levels (0,1). Thanks
@@uruguayanmama Then just allow the two category variable to be "continuous" because practically it makes it no difference if it's only two levels/groups.
Thank you so much Dr. Erin! It was really helpful! About the coding options - if I have an independent variable of 3 groups, can I choose another coding method (not Indicator) that will make all the possible comparisons (1-2, 1-3, 2-3)?
You can't do them all at once in one analysis (it's a math thing, which I can explain if you want). You would just need to run it twice to get 1-2, 1-3, and then separately 2-1, 2-3.
Generally, you mean center continuous variables (i.e. each score minus the grand mean for that variable) to help with interpretation of the coefficients. You wouldn't do that here because the moderator is categorical.
As far as I know it should automatically provide them as long as the interaction meets the pvalue you set in the "probe interaction" section. If you are not seeing it in the output that's why.
@@StatisticsofDOOM I figured this may be it! I switched it to probe all and I found it there. One was significant, two trending, and one non-sig but the unconditional interaction was non significant. So now I’m trying to figure out how to interpret this hah.
Hi Dr. Buchanan, Sorry to keep bothering you here - but it seems like the way you explain statistics is one of the few ways that makes sense to me. I don't mean to burden your RUclips inbox, but I'm hoping you can answer a quick question for me. I'm wondering how to interpret a non-significant unconditional effect, yet finding a significant conditional effect once the interactions are "always probed." I think the unconditional effect shows that the variable as a whole is not a mediator; however, there is a mediating effect at the specific conditional levels that shows that one of the conditions has a significant moderation. I've heard that the unconditional effect is an omnibus test, and thus the only time you can look at the simple slopes in the conditional effects would be if the unconditional effect is significant. But I'm wondering if there is a way to argue moderation when the unconditional effect is non-significant yet I find one significant condition when testing the conditional effects? I really appreciate you taking the time to make these videos. They have been one of the most crucial resources for me to survive my PhD program.
@@DullRio Careful not to mix terms between mediation and moderation. If the overall interaction is not significant, then there is not an interaction, regardless of the simple slopes.
Thank you so much for the video. Dr. Erin, how would you analyze IV as a dichotomous, with a categorical moderator and a second continuous moderator? My DV has also 3 levels (0, 1 and 2) I really appreciate your help.
In a perfect world, sure, but as long as the imbalance isn't super high (like 10 to 1) and the smaller group is not very small N values, then it should be ok.
@@StatisticsofDOOM Thank you very much for your answer. One last question, if I have an independent variable and a dependent variable that are individuals' scores on scales, could variable M be a characteristic of the organization they work for (e.g. public or private, low or high performance)? That is, can individual scores be IV and DB, and one moderator an organizational characteristic?
Thank you for your great video! I have a question: If I have a categorical X (0,1,2) , a categorical moderator (0,1) , a continous Y , and a categorical covariate (0,1). I want to see if there is a moderation effect. However, my Y doesn't conform to normal distribution( so I think I can't use two-way ANCOVA..?). Then what method should I use instead? Can I still use PROCESS or is there any other method I can use? Thank you!
Dear Dr.Erin, thank you so much for your videos helped me a lot. If I have moderator as dummy coded: 0 & 1(male/female) and IV is continuous. Should I choose mean center my IV or both?
Hi! great video, I am wondering if I am testing the interaction with a continuous IV, continuous DV, and categorical Moderator with 4 levels-- can PROCESS 4.1 automatically contrast code my moderator with 4 levels through the multicategorical option using effect coding, or do I need to first manually create 4 contrast coded categories (0, 1, -1) before running it in PROCESS?
If the p value for the interaction is not below a certain value, you will not see the conditional effects. Try changing that value or "probe" interactions so you can see them.
Amazingly helpful video, thank you so much! I am wondering how to check that the regression assumptions are met if I have a categorical IV (4 conditions), a continuous moderator, and continuous DV? I am also using HAYES model 1, except with a categorical X. Would I do the same as in your video, except use the continuous moderator as opposed to continuous X to check for outliers, homoscedasticity, linearity, and normality ? Kind regards :)
Hi Erin, Thank you so much for the videos you have posted on moderation, they have really helped me with my project. I am just wondering why 5 predictors were entered into Gpower? because I thought one of those variables was the DV. Also if you happen to have a distribution that's non-normal (this is what I have) would it make sense to use bootstrapping for the moderation?
Power was estimated based on the number of predictors - but here we have a categorical M variable, so you have to account for how many predictors that creates. The second answer depends on what is non-normal, but bootstrapping is always a good thing.
@@StatisticsofDOOM Amazing thanks Erin!!! Sorry for the late reply, my project went really well and your videos were a massive help. Thank you, thank you, thank you!!😊
Hello, thank you so much for this video. I was doing well until I had to run the regression and it said that I need 3 levels for my moderator (mine has two as its gender, male and female). Is there another analysis I should be carrying out? I am looking at how catholic collective narcissism predicts sexism moderated by gender. Thank you!!!
Hi! I was following your video while I was trying to do it myself and it was going really well until i got to the simple slopes part, my output didn't show anything for this part... I did everything like you did but I noticed when I got to that part in your video it says "conditional effects of the focal predictor at values of the moderator(s)", I had nothing like this on mine. Do you have any idea what I could be doing wrong? Thank you in advance!
Two different possibilities: 1) the version of process you are using, 2) if you don't see the interaction part at all, that means it was not significant, so it doesn't show you the output. You can tell it to show you non sig values by changing the p-value setting in the options.
@@brendaemviagem got the same problem. There´s one significant interaction and one is not significant. First I couldn´t see "conditional effects of the focal predictor at values of the moderator(s)". Then I changed the option of "heteroscedasticity-consistent inference" from "H4" to "none". Maybe that´s helpful. :)
Dear Erin, Thanks for your step-by-step video! May I ask a question? Now I have a dichotomous X (coded as 0&1), and with One-way ANOVA and Linear model, this X is significant to affect Y. But with PROCESS Macro model 1, the main effect of X is not there. Is that because PROCESS Macro model 1 cannot accept a dichotomous X? Thank you very much if you could reply!
It can do dichotomous X as far as I am aware - there's a special button for it. (or is that just M? I can't remember and don't have SPSS). It shouldn't matter though because that test should be the same mathematically. How different are they?
Hello, Thank you for the explanation! I was wondering if it is also possible to do this with a moderator and predictor(IV) as categorical variables with dummy coding, or if it is better to run an ANCOVA instead? Kind regards, Paloma
What if we have continuous moderators, two IVs and one the IVs is within/repeated and the other one is between? In sum, what should we do when we have a mixed design with continous moderators? Thanks.
The interaction doesn't mean that individual predictor is significant - it means that the b values for that predictor are different at different levels of the other predictor. So at low M, you might have .12, while at high M you might have .07 which are different from each other statistically, but neither actually predict Y. You have to decide if that's important enough say that moderation happened.
Hi! Thanks so much for you video! I have a question regarding the multicategorical comparisons in PROCESS: My moderator variable W is categorical (1-4), signifiying my four different groups of patients. When I choose te "indicator" comparison in PROCESS, I will only see the comparisons of the groups to the first group, right? Could you explain to me how I can see the other comparisons as well (so for example group 2 to 3, group 2 to 4, group 3 to 4)? Thank you :)
You are correct - you will need to just reorder the categories and rerun the analysis to see the other comparisons. So, you can create a variable where the 1 group becomes 4, 2 becomes 1, 3 becomes 2, etc. so they are shuffled around until you see all the comparisons.
Hello, I´m doing a moderation study and I need to calculate the sample, I don´t have any table like you have... what steps should I take? and what do I have to put in squared multiple correlation p? Thank you
Hi, as I was following along, I have noticed that Process 4.0 version may not produce output that includes Conditional effects? Is this the case? If so, how can we interpret without conditional effects?
If you aren't seeing them, it's likely because the p threshold you set was not reached (i.e., there's an option to only show them if p < .05 for example). Try changing that value to be higher if you need to see them, but remember that means that they are not be useful if they don't reach your alpha criterion.
@@chenjing9300 The central limit theorem does start to allow for normal distributions at around N ~ 30, but in general I would say that N = 40 is a fairly small sample for a robust test of hypotheses, especially for moderation analysis.
@@StatisticsofDOOM Actually, I have two groups of participants (20 participants for each group), and I have two measurements: one is about brain function and the other is about behavioral performance. The brain function is associated with behavioral performance for one group of participants (measured by Pearson's correlation, p0.05)). In order to determine whether the association between brain function and behavioral performance are different for the two groups, I can (1) carry out mediation analysis to see whether the interaction effect is significant or (2) directly compare the Pearson's correlation coefficients and see whether they are significantly different. The results of these two measures are a bit different. Which one do you think is better? should I use the second since the sample size is not large enough? Many thanks for your time!
@@StatisticsofDOOM Sorry I have one more question, is it necessary to first use the "split" function of SPSS according to Categorical M (for example if I have two groups of participants: patients vs. controls), and then check the outliers for each group? Does this make more sense because patients and controls are from very different distributions? thanks a lot for your time!
Thank you so much for this helpful video for my Masters thesis! I was just wondering how I should interpret the output in SPSS when my independent variable is multicategorical? Because there are 2 different interactions showing in my output and I don't know how what those mean... Greetings from Belgium!
Since the output is dummy coded, I'd suggest learning more about that: ruclips.net/video/Zv19sslm-S4/видео.html - this video is in R, but you can learn what it is and how to interpret the coefficients.
I'm still a bit confused. Everything made sense but the statistics (p-value, coefficient, confidence intervals) for what is supposed to be just the main effect of X on Y is the same as the statistics for when we look at the simple slope of the group coded as 1. So is the row with the main effect of X on Y ONLY the statistics for the specific reference group? How can we get the main effect of X on Y independent of the moderator?
How simple slope results can be different from linear regression made in each category of the moderator made by split file? As I understood simple slope results show linear regression made in each category, but in my case p values are different with each analysis. Did I get it wrong?
@@StatisticsofDOOM I get that. But if simple slope is a regression of the DV on the IV at a specific category of the moderator, why p values differ from simple linear regression results made in each category of the moderator (e.g., age group)? Simple slope shows that there is a significant relationship in a certain age group while linear regression analysis shows that the relationship is insignificant in that group. Or is simple slope not a regression?
@@HereToStay117 They are both regressions, but on different data. The overall model has all the data, while the simple slopes have only one of the categories, which naturally makes them different.
@@aliciaeke895 Many people treat those types of variables as continuous enough ... however, if you want to truly use them as ordinal, you would not be able to use process.
I have two categorical IVs (both dummy coded: 0 & 1), I used both PROCESS Macro (Model 1) and ANOVA in SPSS, but the results are different. Could you explain why is that?
more details in my case for clarification - I guess my question is, why I get different "main effect" results using ANOVA and PROCESS. The interaction result is the same (same F(t), p...). But results for "main effect" of the two IVs are different. I guess the problem is we cannot use PROCESS model 1 to interpret main effect? Since it is a model to test moderation?
@@xiaohanhu5272 Potentially this is the way that the sum of squares is treated in anova and process. If you have completely categorical data, ANOVA is likely a bit easier to run and interpret.
Hi, Thank you, I'm conducting moderation with a categorical dependent variable. So, first I need to dummy code my DV behaviour intentions (intentions to sign up a child for sports program) which has a 5-point Likert scale. However, I was wondering whether someone knows articles which did the same analysis? Regards, Lianne
@@StatisticsofDOOM My DV is a 5-point Lickert scale ranging from 1 (not at all) to 5 (very much) so can I treat it as a continuous variable or do I need to recode it (dummy)? I read in an article when the DV has 5-points I can treat it as continuous, by centring the mean.
@@StatisticsofDOOM Thank you! question on mediation analysis- my supervisor didn't use PROCESS. The Sobel test from Preacher and Kelley (2011), is eliminated from the 3.3 PROCESS version. From what I understand, there's a mathematical error in the formula (Wen and Fan, 2015, Psychological Methods). However, the Sobel test did explain the significance level of the model with mediation and therefore how do I explain the mediation model without this test? The article says use other statistical information, I've no clue which statistical test to use. Many thanks
Sorry to bother you again. I found the answer, in your other video's. However, I was wondering, how to write up a logistic regression (DV= -1 and +1) in the v3.3 output. Basically, the step x variable predicting y, ignoring the mediation (total effect model) is missing? I might be very blind, or I've looked way too many hours on the screen but I cannot figure it out. Do you have a sample write up for logistic regression with mediation analysis?
Great video, thank you so much!!
Thanks for the kind words.
Dear Erin, thank you so much! Your videos helped me a lot with my master’s thesis.
Glad to be of help!
Thanks. This was extremely helpful.
You are amazing...can you run for president?? Really. America needs you.
Ha! Thanks for the kind words. Some real numbers folks would be helpful up in Congress ;).
Thank you so much! Your videos helped me a lot with my bachelor’s thesis! ♡
Thank you for the kind words!
This video is just what I was looking for! Thank you!
Glad to be of help!
Thank you so much for this! This was incredibly helpful for my Masters thesis! :)
Thanks for the kind words!
There are very few videos on RUclips that are just as useful. You are awesome! I just wondered. If you want to compare also average and high groups, do you need one more analysis that covers just these groups?
Nah - I would calculate the confidence interval of the slope and see if those two confidence intervals overlap if you want to show they are different.
Great
In my case I have a moderator with four levels (four different experimental conditions) that are independent to each other, but the program gives to me W1, W2,W3, instead of W1, W2,W3,W4, why? thank you!
Check out some videos on dummy coding (I have a few on the channel), you end up with w4 versus w1, w4 versus w2, and w4 versus w3 as your predictors.
Dear Dr. Erin,
Thank you for your video.
Given that referece group is a low group, I think the results do not indicate whether the slope of average group (m2) is significantly different from the slope of high group (m3). How can we compare the slopes of m2 and m3? Do we have to reanaylze after chaging the reference group?
Exactly correct!
Fantastic, thank you:)
Hello there, thanks for the helpful video! I have a problem regarding the output SPSS gives me. For some reason that I cannot identify I do not get the output on the "conditional effects of the focal predictor at values of the moderator". Instead the output just stops at the "test for highest order unconditional interaction". Have you ever faced a similar issue or suspect a reason? Thanks!
That implies that the p value for the interaction is not significant. There's an option to show those effects at a higher p value, which you need to change to if you want to see it when non-significant.
Dear Erin, thank you so much for your videos - they are super helpful! A quick question please, what if my X is categorical with 2 levels high vs. low, my M is 2 levels as well with high vs. low and my Y is two different types of products (hedonic vs. utilitarian), would I still be able to use simple moderation 1 and dummy code all my 3 variables, or should I use ANOVA, or any other analysis? Thanks so much for your response in advance!
That's probably a multiway frequency analysis given than all the variables are categorical!
I think it's helpful, but still figuring out how exactly to do it... I have a moderator of 4 different categories (without order). How would I do that? Same as you and code the last one as 1 & 1?
The new version of process has categorical moderators I believe, you just have to indicate it's categorical so you get simple slopes by group, rather than low, average, and high.
Dear Dr.Erin, thank you so much for your video！Do I need to encode(e.g., Indicator) both the independent variable and the moderator variable when both are categorical variables with three groups, and the dependent variable is continuous? If so, what specific metrics(e.g., X1*W1) should I look at in the subsequent analysis?
Yes you would need to encode both - I’m not sure about the newer versions of process but it should handle that!
Thanks a lot for your videos! I am stuck with a model where I have a categorical IV (more than two levels), one continuous moderator, and a continuous DV. Can I use PROCESS to run such an analysis? If I use ANOVA with a mean-centered moderator, I think I would only be able to talk about the effect at the average value of the moderator, unlike PROCESS where I am able to talk about the effect 1SD above and below the average score of the moderator (which is a pretty nice thing I have learned from you). What would be your suggestion? Hope you have a great 2022!
If you are using the newer version of process, I believe so - I'm not in this video but you can use categorical M in many new places, so try it and see what happens!
Hi. Thanks so much for this video. I noticed that your categorical variable has 3 levels. Is there a problem using a categorical variable that has only two levels (0,1). Thanks
Nope - that's actually much easier! :)
@@StatisticsofDOOM Process asks for at least three categories to run it like you show in the video...
@@uruguayanmama Then just allow the two category variable to be "continuous" because practically it makes it no difference if it's only two levels/groups.
Hi Erin. What would you suggest for a similar model but with a categorical variable that has only 2 levels?
Same model, easier to interpret because you’ll only get one dummy coded variable.
Thank you so much Dr. Erin! It was really helpful!
About the coding options - if I have an independent variable of 3 groups, can I choose another coding method (not Indicator) that will make all the possible comparisons (1-2, 1-3, 2-3)?
You can't do them all at once in one analysis (it's a math thing, which I can explain if you want). You would just need to run it twice to get 1-2, 1-3, and then separately 2-1, 2-3.
"The following variables were mean centered prior to analysis" what does it mean? what should i do different? thank you very much for the video.
Generally, you mean center continuous variables (i.e. each score minus the grand mean for that variable) to help with interpretation of the coefficients. You wouldn't do that here because the moderator is categorical.
These videos are always so helpful. I'm wondering how to get the simple slopes for v3.5, my moderator is 4 levels
As far as I know it should automatically provide them as long as the interaction meets the pvalue you set in the "probe interaction" section. If you are not seeing it in the output that's why.
@@StatisticsofDOOM I figured this may be it! I switched it to probe all and I found it there. One was significant, two trending, and one non-sig but the unconditional interaction was non significant.
So now I’m trying to figure out how to interpret this hah.
Hi Dr. Buchanan,
Sorry to keep bothering you here - but it seems like the way you explain statistics is one of the few ways that makes sense to me. I don't mean to burden your RUclips inbox, but I'm hoping you can answer a quick question for me.
I'm wondering how to interpret a non-significant unconditional effect, yet finding a significant conditional effect once the interactions are "always probed." I think the unconditional effect shows that the variable as a whole is not a mediator; however, there is a mediating effect at the specific conditional levels that shows that one of the conditions has a significant moderation.
I've heard that the unconditional effect is an omnibus test, and thus the only time you can look at the simple slopes in the conditional effects would be if the unconditional effect is significant. But I'm wondering if there is a way to argue moderation when the unconditional effect is non-significant yet I find one significant condition when testing the conditional effects?
I really appreciate you taking the time to make these videos. They have been one of the most crucial resources for me to survive my PhD program.
@@DullRio Careful not to mix terms between mediation and moderation. If the overall interaction is not significant, then there is not an interaction, regardless of the simple slopes.
Thank you so much for the video. Dr. Erin, how would you analyze IV as a dichotomous, with a categorical moderator and a second continuous moderator? My DV has also 3 levels (0, 1 and 2)
I really appreciate your help.
I believe you just need to mark the categorical moderator using the process options for categorical coding.
Thanks for the video, it's very good. I have a question: should the groups or the categorical variable have similar amounts of data?
In a perfect world, sure, but as long as the imbalance isn't super high (like 10 to 1) and the smaller group is not very small N values, then it should be ok.
@@StatisticsofDOOM Thank you very much for your answer. One last question, if I have an independent variable and a dependent variable that are individuals' scores on scales, could variable M be a characteristic of the organization they work for (e.g. public or private, low or high performance)? That is, can individual scores be IV and DB, and one moderator an organizational characteristic?
@@cristianoyarzun9084 Sure!
Thank you for your great video! I have a question: If I have a categorical X (0,1,2) , a categorical moderator (0,1) , a continous Y , and a categorical covariate (0,1). I want to see if there is a moderation effect. However, my Y doesn't conform to normal distribution( so I think I can't use two-way ANCOVA..?). Then what method should I use instead? Can I still use PROCESS or is there any other method I can use? Thank you!
Not sure I saw this message - depends on what distribution the model does have! If you have a large amount of data, it would be ok to use ANCOVA.
Dear Dr.Erin, thank you so much for your videos helped me a lot.
If I have moderator as dummy coded: 0 & 1(male/female) and IV is continuous. Should I choose mean center my IV or both?
You should mean center the continuous IV but not the categorical variable.
@@StatisticsofDOOM thank you very much ❤️
Hi! great video, I am wondering if I am testing the interaction with a continuous IV, continuous DV, and categorical Moderator with 4 levels-- can PROCESS 4.1 automatically contrast code my moderator with 4 levels through the multicategorical option using effect coding, or do I need to first manually create 4 contrast coded categories (0, 1, -1) before running it in PROCESS?
Sounds like it does the coding for you - just make sure it's the comparisons you actually wish to run.
I have run my moderation analysis like this, but the simple slopes are not appearing in the output. What do I do??
If the p value for the interaction is not below a certain value, you will not see the conditional effects. Try changing that value or "probe" interactions so you can see them.
Amazingly helpful video, thank you so much!
I am wondering how to check that the regression assumptions are met if I have a categorical IV (4 conditions), a continuous moderator, and continuous DV? I am also using HAYES model 1, except with a categorical X.
Would I do the same as in your video, except use the continuous moderator as opposed to continuous X to check for outliers, homoscedasticity, linearity, and normality ?
Kind regards :)
I would check the continuous variables only in that scenario!
@@StatisticsofDOOM thank you!!!!
Hi Erin,
Thank you so much for the videos you have posted on moderation, they have really helped me with my project.
I am just wondering why 5 predictors were entered into Gpower? because I thought one of those variables was the DV. Also if you happen to have a distribution that's non-normal (this is what I have) would it make sense to use bootstrapping for the moderation?
I should have added that my sample is quite large (almost 2,000) so I've just referenced the central limit theorem to say it shouldn't be an issue.
Power was estimated based on the number of predictors - but here we have a categorical M variable, so you have to account for how many predictors that creates. The second answer depends on what is non-normal, but bootstrapping is always a good thing.
@@StatisticsofDOOM Amazing thanks Erin!!! Sorry for the late reply, my project went really well and your videos were a massive help. Thank you, thank you, thank you!!😊
Hello, thank you so much for this video. I was doing well until I had to run the regression and it said that I need 3 levels for my moderator (mine has two as its gender, male and female). Is there another analysis I should be carrying out? I am looking at how catholic collective narcissism predicts sexism moderated by gender. Thank you!!!
You don't need more than two levels mathematically ... not sure if that's a new thing in PROCESS though, as I don't have access to SPSS anymore.
Hi! I was following your video while I was trying to do it myself and it was going really well until i got to the simple slopes part, my output didn't show anything for this part... I did everything like you did but I noticed when I got to that part in your video it says "conditional effects of the focal predictor at values of the moderator(s)", I had nothing like this on mine. Do you have any idea what I could be doing wrong? Thank you in advance!
Two different possibilities: 1) the version of process you are using, 2) if you don't see the interaction part at all, that means it was not significant, so it doesn't show you the output. You can tell it to show you non sig values by changing the p-value setting in the options.
@@StatisticsofDOOM Ohh ok, I think it is the second option then. Thank you very much!
@@brendaemviagem got the same problem. There´s one significant interaction and one is not significant. First I couldn´t see "conditional effects of the focal predictor at values of the moderator(s)". Then I changed the option of "heteroscedasticity-consistent inference" from "H4" to "none". Maybe that´s helpful. :)
@@sophialefin9421 yes, I already did that by now ahah, but thank you so much for helping anyway! ☺️
Dear Erin,
Thanks for your step-by-step video! May I ask a question? Now I have a dichotomous X (coded as 0&1), and with One-way ANOVA and Linear model, this X is significant to affect Y. But with PROCESS Macro model 1, the main effect of X is not there. Is that because PROCESS Macro model 1 cannot accept a dichotomous X?
Thank you very much if you could reply!
It can do dichotomous X as far as I am aware - there's a special button for it. (or is that just M? I can't remember and don't have SPSS). It shouldn't matter though because that test should be the same mathematically. How different are they?
@@StatisticsofDOOM Thanks for your reply Erin. I think it can do dichotomous X as well. I will explore to find the answer.
Hello,
Thank you for the explanation! I was wondering if it is also possible to do this with a moderator and predictor(IV) as categorical variables with dummy coding, or if it is better to run an ANCOVA instead?
Kind regards,
Paloma
Much easier to use ANCOVA when IV and DV are categorical - it actually would reduce down to that, mathematically.
What if we have continuous moderators, two IVs and one the IVs is within/repeated and the other one is between? In sum, what should we do when we have a mixed design with continous moderators? Thanks.
It sounds like you should consider a multilevel model because the moderator has multiple measurements.
Hi Dr, Erin..how should I interpret if I've a significant interaction but it's insignificant for both of my conditional effects? Thanks!
The interaction doesn't mean that individual predictor is significant - it means that the b values for that predictor are different at different levels of the other predictor. So at low M, you might have .12, while at high M you might have .07 which are different from each other statistically, but neither actually predict Y. You have to decide if that's important enough say that moderation happened.
Hi! Thanks so much for you video! I have a question regarding the multicategorical comparisons in PROCESS: My moderator variable W is categorical (1-4), signifiying my four different groups of patients. When I choose te "indicator" comparison in PROCESS, I will only see the comparisons of the groups to the first group, right? Could you explain to me how I can see the other comparisons as well (so for example group 2 to 3, group 2 to 4, group 3 to 4)? Thank you :)
You are correct - you will need to just reorder the categories and rerun the analysis to see the other comparisons. So, you can create a variable where the 1 group becomes 4, 2 becomes 1, 3 becomes 2, etc. so they are shuffled around until you see all the comparisons.
@@StatisticsofDOOM great, thank you so much! :)
Hello, I´m doing a moderation study and I need to calculate the sample, I don´t have any table like you have... what steps should I take? and what do I have to put in squared multiple correlation p? Thank you
Check out g*power to give an idea of what you might do to estimate the sample - it's also shown in this video (not sure what you mean by table).
Hi, as I was following along, I have noticed that Process 4.0 version may not produce output that includes Conditional effects? Is this the case? If so, how can we interpret without conditional effects?
If you aren't seeing them, it's likely because the p threshold you set was not reached (i.e., there's an option to only show them if p < .05 for example). Try changing that value to be higher if you need to see them, but remember that means that they are not be useful if they don't reach your alpha criterion.
What if the assumptions for regressions were not met for moderation analysis, such as normality?
Is the data large enough to assume the sampling distribution is normal? (i.e. N is large?)
how about N=40, is it large enough？
@@chenjing9300 The central limit theorem does start to allow for normal distributions at around N ~ 30, but in general I would say that N = 40 is a fairly small sample for a robust test of hypotheses, especially for moderation analysis.
@@StatisticsofDOOM Actually, I have two groups of participants (20 participants for each group), and I have two measurements: one is about brain function and the other is about behavioral performance. The brain function is associated with behavioral performance for one group of participants (measured by Pearson's correlation, p0.05)). In order to determine whether the association between brain function and behavioral performance are different for the two groups, I can (1) carry out mediation analysis to see whether the interaction effect is significant or (2) directly compare the Pearson's correlation coefficients and see whether they are significantly different. The results of these two measures are a bit different. Which one do you think is better? should I use the second since the sample size is not large enough? Many thanks for your time!
@@StatisticsofDOOM Sorry I have one more question, is it necessary to first use the "split" function of SPSS according to Categorical M (for example if I have two groups of participants: patients vs. controls), and then check the outliers for each group? Does this make more sense because patients and controls are from very different distributions? thanks a lot for your time!
Thank you so much for this helpful video for my Masters thesis! I was just wondering how I should interpret the output in SPSS when my independent variable is multicategorical? Because there are 2 different interactions showing in my output and I don't know how what those mean... Greetings from Belgium!
Since the output is dummy coded, I'd suggest learning more about that: ruclips.net/video/Zv19sslm-S4/видео.html - this video is in R, but you can learn what it is and how to interpret the coefficients.
I'm still a bit confused. Everything made sense but the statistics (p-value, coefficient, confidence intervals) for what is supposed to be just the main effect of X on Y is the same as the statistics for when we look at the simple slope of the group coded as 1. So is the row with the main effect of X on Y ONLY the statistics for the specific reference group? How can we get the main effect of X on Y independent of the moderator?
By setting one group to zero as their coding, they are the main effect and the simple slope. I think this is what you are asking.
@@StatisticsofDOOM So the main effect is ONLY for the group coded as 0, not all groups correct?
@@HugoSH93 Right - you should get the separate ones in the simple slopes output.
How simple slope results can be different from linear regression made in each category of the moderator made by split file? As I understood simple slope results show linear regression made in each category, but in my case p values are different with each analysis. Did I get it wrong?
The p-values for the slopes may be different across groups. Is that what you are asking?
@@StatisticsofDOOM I get that. But if simple slope is a regression of the DV on the IV at a specific category of the moderator, why p values differ from simple linear regression results made in each category of the moderator (e.g., age group)? Simple slope shows that there is a significant relationship in a certain age group while linear regression analysis shows that the relationship is insignificant in that group. Or is simple slope not a regression?
@@HereToStay117 They are both regressions, but on different data. The overall model has all the data, while the simple slopes have only one of the categories, which naturally makes them different.
mam do we have to take those items for moderation which were deleted during EFA due to lower factor loadings?
I am confused - this video is over moderation ... what is the EFA question?
what if independent,dependent and moderator all are second order factor? can help me with moderation effect
If I understand correctly, you would need to create this model using structural equation modeling (not process).
@@StatisticsofDOOM Can you guide me to calculate interaction for higher order factors
@@iyersmita14 I'm still not 100% sure what you are asking but I have many structural equation modeling videos, if that's what you are asking.
can anyone help me pleaseee!! does the same method apply if your M variable is ordinal?
My M is a likert scale (1-6), but I've seen research state that likert scales are ordinal rather than interval
@@aliciaeke895 Many people treat those types of variables as continuous enough ... however, if you want to truly use them as ordinal, you would not be able to use process.
I have two categorical IVs (both dummy coded: 0 & 1), I used both PROCESS Macro (Model 1) and ANOVA in SPSS, but the results are different. Could you explain why is that?
more details in my case for clarification - I guess my question is, why I get different "main effect" results using ANOVA and PROCESS. The interaction result is the same (same F(t), p...). But results for "main effect" of the two IVs are different. I guess the problem is we cannot use PROCESS model 1 to interpret main effect? Since it is a model to test moderation?
@@xiaohanhu5272 Potentially this is the way that the sum of squares is treated in anova and process. If you have completely categorical data, ANOVA is likely a bit easier to run and interpret.
@@StatisticsofDOOM Thank you so much for your response! I will use ANOVA result then to interpret my data. Your videos are really helpful!
Hello,I would like to know whether the templates of process 3.3 has been modified by Hays
Some of them are different yes - I have a few videos on Process 3.
@@StatisticsofDOOM OK,Thank you.How can I get the templates of process 3.3
For that, you will need to buy the book - they are not online to my knowledge :(
Hi,
Thank you, I'm conducting moderation with a categorical dependent variable. So, first I need to dummy code my DV behaviour intentions (intentions to sign up a child for sports program) which has a 5-point Likert scale. However, I was wondering whether someone knows articles which did the same analysis?
Regards,
Lianne
You will likely need to do a log regression if you are going to treat the DV as categorical.
@@StatisticsofDOOM My DV is a 5-point Lickert scale ranging from 1 (not at all) to 5 (very much) so can I treat it as a continuous variable or do I need to recode it (dummy)? I read in an article when the DV has 5-points I can treat it as continuous, by centring the mean.
@@lianneengwerda3947 Yes, often people treat Likert scales as continuous! Then you should be able to just use it as is.
@@StatisticsofDOOM Thank you! question on mediation analysis- my supervisor didn't use PROCESS.
The Sobel test from Preacher and Kelley (2011), is eliminated from the 3.3 PROCESS version. From what I understand, there's a mathematical error in the formula (Wen and Fan, 2015, Psychological Methods). However, the Sobel test did explain the significance level of the model with mediation and therefore how do I explain the mediation model without this test? The article says use other statistical information, I've no clue which statistical test to use.
Many thanks
Sorry to bother you again.
I found the answer, in your other video's. However, I was wondering, how to write up a logistic regression (DV= -1 and +1) in the v3.3 output. Basically, the step x variable predicting y, ignoring the mediation (total effect model) is missing? I might be very blind, or I've looked way too many hours on the screen but I cannot figure it out. Do you have a sample write up for logistic regression with mediation analysis?