00:35 G*Power 3:23 Centring 5:16 Missing data 6:52 outliers 20:10 multicollinearity 21:42 normality, linearity, homogeneity, homoscedasticity 22:30 setting up PROCESS 30:48 interpret PROCESS output 52:15 Create line chart from PROCESS output
This is really HELPFUL. Thanks! Honestly, I know right away if there's gonna be this time choices in the comment section. Then when I scrolled down, here it is I found in the first row, awesome
This is the most amazing thing that I have ever seen. Can you please just do...everything for all of stats...ever? I learned more from this video (for my dissertation no less), than I have ever learned in 4 years of statistics!
Currently doing an assignment in school and the professor didn't really explain how to report a moderation analysis. Watching this video was significantly better than both class and the hours of research that I spent trying to figure things out. So thanks very much; your channel has been and remains to be extremely educational in many of my research endeavors.
Dr Buchanan thank you so much for your work and videos. The detail you provide, the explanations you share are priceless. You are incomparable, and a born and gifted teacher. You help us to really see and really understand. Like the others below you have made such a big big difference to my stat studies and what I can now achieve. Thank you so so much.
THANK YOU. I am a senior-ish academic who should know better, but was struggling to figure out how to plot simple slopes. This video is amazing -- great resource. Your generosity is much appreciated.
This is just so helpful! I am struggling with doing moderation analysis with continuous variable and did not even realize I can do all this using process macro (typically just use the macro for mediation analysis). This video is just so clear in explaining everything you need to do and how to interpret everything you get. Thank you so much for this!
Hi, Erin! First off - thank you so much! This video was unbelievably clear and helpful. I just installed the Process add-on and started playing around with it for my dissertation. Your video was so much more thorough than Andy Field's :) I love him, but you made it all come together with ease...I will definitely subscribe and try to watch some of your other videos. Also, going through all the assumptions first in the video was essential for people who are new to all of this. Thank you again!
Erin, I'm very happy for this video. It help me in the analysis and comprehension of date bank. I'm concluding Master's Degree in Business Management. Thank's again. Sorry for the english I'm Brazilian and i'm development the english language.
Thank you so much for this great video, it helped me a lot :) As a thank you I have a little tip for you: change your Word Font to Courier (the same font is used in SPSS output for process) - it will keep the all the numbers neatly under each other :)
Thank you so much for this video! It is extremely helpful, not only in understanding and interpreting the whole process, but also formatting the graphs :)
Can you have multiple independent variables in Model 1 (not covariates, but variables that interact with the moderator)? If not, is there any way to do this with PROCESS? I have four IVs and one DV and I want to test for age as a moderator of each relationship.
I think you can only do up to two-two way interactions if I remember correctly. You would need to do it manually if you want to do all of them together in one model.
Hi! Just wanted to say THANK YOU! Video was very helpful your responses to questions in the comments section was very helpful! Thank you thank you thank you!!
Statistics of DOOM I do have a question...how are the levels calculated if not just by SD? I did moderation and used the percentage points of 16, 50, and 84%...is that accurate?
@@seanfdaly4 You can do it that way as well, but do realize that's basically the same idea as the SD, since those percentages correspond to 1 SD in the z normal distribution.
Question about the model summary statistics (ie. 31:46)...since this is a moderation regression, with 'books' as the IV, and 'attendance' as the moderator; could you interpret it as: 'books' accounts for 40.22% of the variance in grades when moderated by 'attendance'? Or is this wrong?
Yeah something like that - for the main effect it's basically that it accounts for 40% of the variance controlling for attendance and the interaction of the two.
Thanks Erin for this video! It was really useful. I was wondering if you are going to upload a video on MODERATED MEDIATION with PROCESS, especially one dealing with several IVs. This type of analysis is becoming popular in psychology papers. Thank you!
Hopefully someone reads this but at 34:12, why is that predictor significant with 0.148? Is it because it's less than the constant (a) of 61.6024? Thanks!
I have applied same things which you explained in video. My interaction is Insignificant ( in video you said there is no specific explanation of significance and gave Three level L , M , H. Please guide me literature reference from where you took this explanation.) My supervisor wants to see literature reference as well as book reference. Please give me one more favour that provide reference of explanation of Johnson-Neyman Techniques. If you have used all these in any of your publication or any other. Please accept my advance Thanks :)
Hi Dr. Buchanan. Thank you for breaking that down so simply. Really helped me understand how to interpret the interaction. One question: How would you interpret the output for a dichotomous moderator? Your data has 3 levels of attendance with low, medium and high, where mine has only two; one is the control condition (0) and the other the experimental (1). I am looking at pre- and post-test scores, where the pretest scores are my predictor variable, post-test scores the DV, and the condition(s) the moderator. My output did not include the Johnson-Neyman method because it could not be used with a dichotomous moderator, but it did note that the moderator was mean centered. How does it mean center a dichotomous variable? Also, what if you were to include covariates? Thanks again! Cari
I would turn off mean centering if you use a dichotomous moderator. My variables are continuous, so it created the low, average, and high for me. Mean centering does not make sense for categorical variables. If you have a dichotomous moderator, you can interpret it as the simple slope for each group separately.
@@StatisticsofDOOM Hi Erin, I have a similar question - for my predictors, one is continuous and the other is dichotomous - how would I check assumptions are met for this analysis? Also - this video has helped me immensely, thank you so much!
@@t1cketyboo if it's only dichotomous (meaning only two categories), you can do the data screening like normal and shown in this video - you may see some weird splits in the plots because of the groups, but generally if it's ok assumptions wise, that doesn't affect too much.
Thank you very much for this detailed video, what a great resource. I was particularly confused regarding the use of GPower in this situation, and your video helped a lot. Unfortunately I am still unsure as to how to approach this for my particular situation. I am interested in the interaction between a categorical IV with 4 levels, and a continuous moderator. Does this still mean I need to enter '3' as the number of predictors (1 IV, 1 moderator, and their interaction) in GPower? Or do I need to count each dummy variable that PROCESS calculates as a predictor and thus see this as using 7 predictors (moderator, D1, D2, D3, int_1, int_2, int_3)? I assume it is the latter, but I am just not entirely sure... I am sorry to ask, but have been unable to find a clear answer to this anywhere...
Hey Erin, thanks for your video. It's helpful to go through it step by step. I've run into a weird problem: doing a linear regression analysis in with 2 predictors + interaction term in spss I find p=.030 for interaction. That's why I'm figuring out PROCESS now to see how the interaction works. However, in PROCESS, I find p=.108 for the interaction. I DO however find conditional interaction effects with 2x p
Something must be different ... maybe try centering your variables before running regular regression? Then try the process on the non centered variables? It should not turn out that different!
If i have got 2 independent variables, 1 dependent variable, and 1 moderating variable, should i add the moderating variable to the independents at 7:05 ? (Yes i have no idea what i am doing)
Hi Erin, what are some steps for troubleshooting bent and crooked lines in the line graph? For some reason SPSS includes 0 as a value on my X-axis! Thanks.
Hey, I've got a question: At the 'Conditional effect of X on Y at values of the moderator' part of the output, my p-values are all insignificant, what should I do. Also, there were no significant direct effects, but my interaction effect is significant (p = .0492)
Hi Erin, thanks for this great video! I had a quick question about determining the direction of moderation. If the interaction is significant (and a positive coefficient) but the conditional effects are negative numbers, and also significant, do we rely on the conditional effects when determining the direction of the moderation?
You would only look at the simple slopes to explain the effect, as the interaction coefficient has the other variable also included (and that coefficient isn't really interpretable anyway, except that it is "not zero").
Hi Dr. Buchanan, love your videos. They are really clear and helpful. And really liked the simple slopes in this vid, but some of the mediation analyses may be misleading. I've been struggling with interpreting mediation results using Process. Particularly, I think I was wrongly using the causal steps approach to mediation (i.e., significant a path, b path, c path and c' path) to interpret Process' bootstrapping results (i.e., estimating confidence intervals). However, Hayes (2014) cautions against using the casual steps approach for several reasons (assumptions of normality, weak power, "inconsistent" mediation per Baron and Kenny, etc...) and cautions against using the outdated terms of full and partial mediation. See Hayes (2014), section 6.1 What do you think?
I don't disagree. There are lots of approaches to mediation. I propose presenting your results in the most understandable way possible, which would allow for others to decide if they believe the evidence you are presenting. Additionally, I would not use the word "cause" without a supporting experimental research design (which is a different issue than the question of mediation).
Dr. Buchanan, Thank you so much for this video. It has been extremely helpful. I am writing up the results for my moderation model and I was wondering if you had citations for the process in which you screened and removed any leveraging outliers prior to creating running the moderation. Any information would be helpful. thank you.
Thanks Erin, great video. I need some clarification on the POWER that you touched upon earlier in the video. As far as predictors (K) are concerned, if I have one IV, one moderator, 1 mediator, and 1 DV, what is the total number of predictors? In other words, sample size required for the medium effect size and power .8. Thanks for your help...
Hi this video is extremely helpful but I have a question, what is the values in the coefficients column are negative, what do they imply? Thank you so much!
Hi Erin, when explaining how to report the interaction, you interpret coeff as b (standardized coefficients) while (correct me if I'm wrong) they are unstandardized coefficients in fact.I would like to be able to interpret the strength of the predictor (X) and also see if it is comparable to the previous findings on this topic. If you could relate to this issue, would be great. Thanks!
Thank you so much Dr. Buchanan, your video is excellent and very very helpful. I have a question that maybe you can help with. I ran moderation and the interaction effect was non-significant, however; in the conditional effect table (high and low groups), all effects become statistically significant. Does this mean I can use this and say moderation does occur for these specific groups?
This was so helpful, thank you! My interaction effect was not significant. Would this be a reason for not seeing the johnson neyman output or the conditional effects? I should add that I am also using the new version3.3
Yes, I think you only get the Johnson Neyman when the interaction is significant. There's a button you can tell it to give you the interaction either way.
Hey Erin! Wonderful video, extremely useful! I do have a quick question though - I have one continuous moderator and one categorical IV with four levels. I am guessing I have to dummy code the variables and then enter those into the model? Could you please throw some light on this? My DVs are continuous too. Thanks!
I have significant X-> Y effect, Johnson-Neyman output, and for simple slopes, significant p-values for high and centered average of my mediator. However, my interaction was not significant. What does this mean, and can I still interpret my simple slopes?
i have similar results. The independent regression coefficients are significant but the interaction is non significant. Can i interpret moderation still? as some books say that moderation is actually the interaction between the IV and Moderator.
That means that there is main effect of X to Y, but the slopes do not different across M. I would not interpret the simple slopes because the interaction is not significant.
I would not interpret the simple slopes if the interaction coefficient is not significant, as that means all the simple slopes are statistically equal (given power, of course) ... the interaction is between X and M predicting Y, correct, which is how it is covered in this video.
Great video! I can't get the "Conditional effect of X on Y at values of the moderator" output in version 3.5 of PROCESS? Can someone help me on how to get this so I can measure the slopes like in the video?
Hi Erin, I have a question regarding the interaction term in the output. Do I have to include it in my write up and what does it mean if the overall moderation model is significant but the interaction term is not? Thank you very much in advance, Max.
Hi Erin, Thanks for the videos. They are all very helpful! I was wondering whether you were planning to do more videos on PROCESS, especially on how to use it when you have an indip. variable with 3 categories (moderator and dependent variable continuous). Looking in the internet it seems that many people are asking the same question. Plus how do you to report PROCESS results in APA style? Any help would be much appreciate it. Thanks!
Hi Erin! Thanks for all your videos that helped me a lot! But just one question: Does there actually a graphical user interface for PROCESS exist? Or should I better use SmartPLS or AMOS for this? The problem is that with these programs it is not possible to calculate with moderations.... only mediations are possible (as far as I know)! :((
I am sorry for being imprecise, but I meant whether it is possible to "draw" / visualise your calculated model with PROCESS. Like it is in AMOS (you have the regressions of the SEM AND the graphical model with its arrows and boxes). Have a nice holiday.... ;-)
No on the graphic interface to Process. I think you might be able to coerce AMOS into interactions, but you would have to make the interaction column manually, and then interpreting would be much harder than the regression way (also df is treated very differently).
I really like your video. It helps me a lot. There are still some question remaining though: 1. How do i identify non-significance? Just by looking at the p-value? Or is the LLCI and ULCI also important? 2. What if my overall Effect is non significant, but one of my moderation effet in the slopes is significant? Am i allowed to predict a relationship between my X on Y in that case? I´d be glad for some help. Thank you!
1) you could do either depending on your area's focus on p values, I would say p values are more comment. 2) that would depend on if the interaction was significant - in general main effects and simple slopes are not interpreted the same.
Thank you Erin, your answer helped me a lot. However when analysing my last Hypothesis i stumbled across another problem: The moderator is gender. My interaction is non significant (p = .95), but one of my two predictors is significant (p
Hello Dr. Buchanan- I will echo the sentiment of many others here and thank you deeply for sharing this video. It has been crucial in the process of proposing and defending my dissertation. I have one question- In the initial screen when opening up the process macro, under "options", I have a drop-down menu for "Heteroscedasticity-consistent inference", rather than a checkbox for "Heteroscedasticity-consistent SEs" like in your video. This drop down menu presents several options such as HCO, HC1, HC2, HC3, HC4, as well as "none" (some of these different HC's also have names in parentheses afterwards, I'm assuming credited to different researchers). Do you have any insight about which one of these options I should choose? Thanks!
Let me refer you to Hayes' publication on this topic: link.springer.com/content/pdf/10.3758/BF03192961.pdf Specifically I find this paragraph useful: For small sample sizes, the standard errors from HC0 are quite biased, usually downward, and this results in overly liberal inferences in regression models (see, e.g., Bera, Suprayitno, & Premaratne, 2002; Chesher & Jewitt, 1987; Cribari-Neto, Ferrari, & Cordeiro, 2000; Cribari-Neto & Zarkos, 2001; Furno, 1996). But HC0 is a consistent estimator when the errors are heteroskedastic; that is, the bias shrinks with increasing sample size. Three alternative estimators, HC1, HC2, and HC3, are all asymptotically equivalent to HC0 but have far superior small sample properties relative to HC0 (Long & Ervin, 2000; MacKin-non & White, 1985). A newer estimator, HC4, is preferred when there are cases with high leverage.
Hi. Thank yuou for your video. I have a problem and if you answer ı would be pleased. My p value in model summary is significant however interaction is not significant. So how can ı interpret it? I know there is no mderation but can ı say that x effects y directly from looking at model summary? Thank you.
Hi, amazing video - really helped me with many of my projects so far! Quick question regarding the outlier "analysis": Is there a reference I can use (as in research paper I can quote) that suggests the 2/3 rule for Mahalanobis, Cooks, and Leverage? Personally, I have only found academic articles that suggest to either use only one or maybe two of them. Thanks in advance for any help! :)
I know people have cited my videos, but for academic sources, I usually use Tabachnick and Fidell's book along with Cohen et al.'s book as citations for the outliers. I usually also state that I use 2/3 to adequately cover the different properties of regression (leverage, discrepancy, and patterns for Mahalanobis) without being too sensitive by only using one.
Hei Erin Tank you so much for this video. It was very helpfull. One question tho. When we tried to make the graph and typed in low, average and high i variable view, it didnt show in dataview when we typed in -1, 0 and 1. It this some kind of setting, or do you think we did something wrong?
Hello Dr. Buchanan. Thank you so much for posting this excellent video! It is very clear and easy to follow. I just wanted to make sure I am understanding my results correctly. My overall model was significant, as was my predictor variable (X) and the moderator (M), however, the interaction itself was not significant. Is it accurate to state the predictor variable (X) and the moderator variable (M) predict Y separately? Also, because the interaction is not significant, it is not appropriate to interpret the simple slopes or the J-N output? Thank you in advance for your guidance!
Exactly what you said - x and m predict separately in the main effects, but there is not an interaction. I would not interpret the JN or the simple slope output because they are follow up tests (like ANOVA).
Hi thank you so for this amazing video, just had a question if in multiple regression (step wise) the IV is not a significant predictor can we still use it in moderation analysis and get significant moderation effects?
Hi, thank you for the video, it's great! When calculating a sample size testing more than one moderator, how many predictors do you put in? (I am testing 3 moderators on a linear relationship)
00:35 G*Power
3:23 Centring
5:16 Missing data
6:52 outliers
20:10 multicollinearity
21:42 normality, linearity, homogeneity, homoscedasticity
22:30 setting up PROCESS
30:48 interpret PROCESS output
52:15 Create line chart from PROCESS output
This is really HELPFUL. Thanks!
Honestly, I know right away if there's gonna be this time choices in the comment section. Then when I scrolled down, here it is I found in the first row, awesome
This is the most amazing thing that I have ever seen. Can you please just do...everything for all of stats...ever? I learned more from this video (for my dissertation no less), than I have ever learned in 4 years of statistics!
Glad to be of help! :)
What an amazing video ! Your teaching is so useful, practical and easily understandable. Thank you so much !!!
Thank you for the kind words!
Currently doing an assignment in school and the professor didn't really explain how to report a moderation analysis. Watching this video was significantly better than both class and the hours of research that I spent trying to figure things out. So thanks very much; your channel has been and remains to be extremely educational in many of my research endeavors.
Thanks for the kind words!
Dr. Buchanan, this is just a little note to express my gratitude for all your excellent tutorials! They are hugely appreciated. Thank you!
Thanks for the kind words!
Dr Buchanan thank you so much for your work and videos. The detail you provide, the explanations you share are priceless. You are incomparable, and a born and gifted teacher. You help us to really see and really understand. Like the others below you have made such a big big difference to my stat studies and what I can now achieve. Thank you so so much.
Thank you for the kind message!
This video is awesome. Great narration and video quality! It is truly one of the best educational videos that I have ever seen at RUclips.
Thank you! You explained something I thought I’d never understood so clearly. I think my report might actually make sense.
Thanks for the kind words! Glad it helped.
THANK YOU. I am a senior-ish academic who should know better, but was struggling to figure out how to plot simple slopes. This video is amazing -- great resource. Your generosity is much appreciated.
You are an amazing lecturer. Extremely thorough, but eloquent at the same time.
Thank you!
This is just so helpful! I am struggling with doing moderation analysis with continuous variable and did not even realize I can do all this using process macro (typically just use the macro for mediation analysis). This video is just so clear in explaining everything you need to do and how to interpret everything you get. Thank you so much for this!
Thanks for the kind words!
Thank you Dr. Buchanan, this video was very helpful. I wish stat professors actually taught like you.
Thanks for this video. I was having such a hard time with moderation, especially the PROCESS plug-in. Really excited to watch your other videos!
Really thanks. I got a lot from it which is the best video for moderation in process in RUclips.
Glad to be of help!
amazing presentation of hard-core statistics! wish I could know your channel early!
Thank you!
Excellent video, with easy to understand interpretation.
Thank you!
You've just saved my dissertation, I think I'm in love with you
Glad to be of help!
Thanks Dr. Erin, your explanation here was super easy to understand and helped greatly with one of my postgraduate assignments!
I'm glad it was a help!
Hi, Erin! First off - thank you so much! This video was unbelievably clear and helpful. I just installed the Process add-on and started playing around with it for my dissertation. Your video was so much more thorough than Andy Field's :) I love him, but you made it all come together with ease...I will definitely subscribe and try to watch some of your other videos. Also, going through all the assumptions first in the video was essential for people who are new to all of this. Thank you again!
love your voice and your clear, step-by-step explanation!!!
Thank you Dr. Buchanan! Your videos are helping me tremendously with my Master's thesis analyses!
Glad to be of help.
Erin, I'm very happy for this video. It help me in the analysis and comprehension of date bank. I'm concluding Master's Degree in Business Management. Thank's again. Sorry for the english I'm Brazilian and i'm development the english language.
Thank you so much for this great video, it helped me a lot :) As a thank you I have a little tip for you: change your Word Font to Courier (the same font is used in SPSS output for process) - it will keep the all the numbers neatly under each other :)
Good to know ha! I usually put things into Excel, but Word can be a bit easier to read for these examples.
EXTREMELY INFORMATIVE VIDEO!!!!!
Good work explaining what everything does in the PROCESS dialog and output boxes!
Thank you very much Erin.
best best best video out here.
Hey there ! Just a piece of advice- you can select for the names to be run as long names if you go to "long names" option :)
That's a newer option! Sometimes it still doesn't run, so I try to go with short names as much as I can.
Thanks a lot for this video. Your teaching is great and you sound super sympathetic.
Thank you!
I could cry, this is so helpful for my thesis paper, thank you!
Glad it was helpful!
This is the most informative video, thank you so much you have just saved my Honors Thesis & my soul!
Thank you so much for this video! It is extremely helpful, not only in understanding and interpreting the whole process, but also formatting the graphs :)
You are a Godsent! Thank you so much for this video, helped me so much! ☺
Thanks for the kind words!
You are an honest to God actual angel. THANK YOU SO MUCH.
Outstanding lecture, this has helped me enormously with my thesis
Thank you!
JOB WELL DONE, that was amazing, you cannot even imagine how much you helped me, lets just say you cannot even measure in SPSS :D, Stay blessed
Thanks for the kind words!
I really like your way of making things simpler to report. Thank you
This was an incredible help - THANK YOU for your clear explanations, and for taking the time to educate us all!
hello i'm writing from Turkey.. Thank you for the video... ı used your way my thesis...
Can you have multiple independent variables in Model 1 (not covariates, but variables that interact with the moderator)? If not, is there any way to do this with PROCESS? I have four IVs and one DV and I want to test for age as a moderator of each relationship.
I think you can only do up to two-two way interactions if I remember correctly. You would need to do it manually if you want to do all of them together in one model.
Thank you so much, you safed my live respectively my bachelor thesis!
Glad to be of help!
Marvelous, thank you, full context, with reporting advices etc.
Hi! Just wanted to say THANK YOU! Video was very helpful your responses to questions in the comments section was very helpful! Thank you thank you thank you!!
Thank for your video! It helps me so much on my assignment!!!
Thanks for the kind words!
Thank you for making sure I didn't look like an idiot in front of my thesis committee member when we had an analysis consultation today
Excellent! :)
Thanks for sharing. It helped me a lot. Keep up the good work.
Thanks!
I'm here to say thank you ! This really helps
Glad to be of help!
This lesson was incredibly helpful and clear. Thank you for uploading this!
Thanks for the kind words!
Thank you!!! This was simple and easy to understand
Thanks! Appreciate the kind words.
Statistics of DOOM I do have a question...how are the levels calculated if not just by SD? I did moderation and used the percentage points of 16, 50, and 84%...is that accurate?
@@seanfdaly4 You can do it that way as well, but do realize that's basically the same idea as the SD, since those percentages correspond to 1 SD in the z normal distribution.
Question about the model summary statistics (ie. 31:46)...since this is a moderation regression, with 'books' as the IV, and 'attendance' as the moderator; could you interpret it as: 'books' accounts for 40.22% of the variance in grades when moderated by 'attendance'? Or is this wrong?
Yeah something like that - for the main effect it's basically that it accounts for 40% of the variance controlling for attendance and the interaction of the two.
Thanks Erin for this video! It was really useful.
I was wondering if you are going to upload a video on MODERATED MEDIATION with PROCESS, especially one dealing with several IVs. This type of analysis is becoming popular in psychology papers. Thank you!
Thank you so much for such nice explanation and going through the important steps!
This video was so helpful, clear and understandable. Thank you so much! Do you have a video on interpreting output with multiple moderators? (Model 2)
You are truly the best.
Thank you for the video. It was very helpful...
Hopefully someone reads this but at 34:12, why is that predictor significant with 0.148? Is it because it's less than the constant (a) of 61.6024? Thanks!
It is significant if you use p < .05 because the p value is .0148 (you are missing the decimal placement).
@@StatisticsofDOOM Thank you, it was a silly question lol
Thank you so much!!! Very helpful 🙏🏻
It's amazing, I learned a lot from this lecture. Thanks for help!
Glad to be of help!
Thank you so much!! This video is amazing, explained everything I needed to know and more!
Thanks for the kind words!
Thank you so much for this video. It really helped fill in that gaps that I really needed to know to finish my research write up :)
Great! Glad to help.
A prayer answered! This is so incredibly helpful! Thank you, thank you, thank you!!!
Thank you so much. You have explained things in a wonderful way.
Glad to be of help!
I have applied same things which you explained in video. My interaction is Insignificant ( in video you said there is no specific explanation of significance and gave Three level L , M , H. Please guide me literature reference from where you took this explanation.) My supervisor wants to see literature reference as well as book reference. Please give me one more favour that provide reference of explanation of Johnson-Neyman Techniques. If you have used all these in any of your publication or any other. Please accept my advance Thanks :)
You'll surely be a part of my graduation speech.
Thank you for the kind words!
Hi Dr. Buchanan. Thank you for breaking that down so simply. Really helped me understand how to interpret the interaction.
One question: How would you interpret the output for a dichotomous moderator? Your data has 3 levels of attendance with low, medium and high, where mine has only two; one is the control condition (0) and the other the experimental (1). I am looking at pre- and post-test scores, where the pretest scores are my predictor variable, post-test scores the DV, and the condition(s) the moderator.
My output did not include the Johnson-Neyman method because it could not be used with a dichotomous moderator, but it did note that the moderator was mean centered. How does it mean center a dichotomous variable?
Also, what if you were to include covariates?
Thanks again!
Cari
I would turn off mean centering if you use a dichotomous moderator. My variables are continuous, so it created the low, average, and high for me. Mean centering does not make sense for categorical variables.
If you have a dichotomous moderator, you can interpret it as the simple slope for each group separately.
@@StatisticsofDOOM Hi Erin, I have a similar question - for my predictors, one is continuous and the other is dichotomous - how would I check assumptions are met for this analysis? Also - this video has helped me immensely, thank you so much!
@@t1cketyboo if it's only dichotomous (meaning only two categories), you can do the data screening like normal and shown in this video - you may see some weird splits in the plots because of the groups, but generally if it's ok assumptions wise, that doesn't affect too much.
Thank you very much for this detailed video, what a great resource. I was particularly confused regarding the use of GPower in this situation, and your video helped a lot. Unfortunately I am still unsure as to how to approach this for my particular situation. I am interested in the interaction between a categorical IV with 4 levels, and a continuous moderator. Does this still mean I need to enter '3' as the number of predictors (1 IV, 1 moderator, and their interaction) in GPower? Or do I need to count each dummy variable that PROCESS calculates as a predictor and thus see this as using 7 predictors (moderator, D1, D2, D3, int_1, int_2, int_3)? I assume it is the latter, but I am just not entirely sure... I am sorry to ask, but have been unable to find a clear answer to this anywhere...
Use each of the dummy coded variables, so 7 yes. :)
I'm sorry for my late response. Thank you so much for your reply! Your channel is a very valuable resource.
Hey Erin, thanks for your video. It's helpful to go through it step by step. I've run into a weird problem: doing a linear regression analysis in with 2 predictors + interaction term in spss I find p=.030 for interaction. That's why I'm figuring out PROCESS now to see how the interaction works. However, in PROCESS, I find p=.108 for the interaction. I DO however find conditional interaction effects with 2x p
Something must be different ... maybe try centering your variables before running regular regression? Then try the process on the non centered variables? It should not turn out that different!
If i have got 2 independent variables, 1 dependent variable, and 1 moderating variable, should i add the moderating variable to the independents at 7:05 ? (Yes i have no idea what i am doing)
Yes, but if you have two two-way interactions because you have two X variables, you will need to use Model 2.
Hi Erin, what are some steps for troubleshooting bent and crooked lines in the line graph? For some reason SPSS includes 0 as a value on my X-axis!
Thanks.
Hey, I've got a question: At the 'Conditional effect of X on Y at values of the moderator' part of the output, my p-values are all insignificant, what should I do.
Also, there were no significant direct effects, but my interaction effect is significant (p = .0492)
Hi Erin, thanks for this great video! I had a quick question about determining the direction of moderation. If the interaction is significant (and a positive coefficient) but the conditional effects are negative numbers, and also significant, do we rely on the conditional effects when determining the direction of the moderation?
You would only look at the simple slopes to explain the effect, as the interaction coefficient has the other variable also included (and that coefficient isn't really interpretable anyway, except that it is "not zero").
@@StatisticsofDOOM amazing, thank you so much for getting back to me! I'd been so confused about this. :)
Hi Dr. Buchanan, love your videos. They are really clear and helpful. And really liked the simple slopes in this vid, but some of the mediation analyses may be misleading.
I've been struggling with interpreting mediation results using Process. Particularly, I think I was wrongly using the causal steps approach to mediation (i.e., significant a path, b path, c path and c' path) to interpret Process' bootstrapping results (i.e., estimating confidence intervals). However, Hayes (2014) cautions against using the casual steps approach for several reasons (assumptions of normality, weak power, "inconsistent" mediation per Baron and Kenny, etc...) and cautions against using the outdated terms of full and partial mediation. See Hayes (2014), section 6.1
What do you think?
I don't disagree. There are lots of approaches to mediation. I propose presenting your results in the most understandable way possible, which would allow for others to decide if they believe the evidence you are presenting. Additionally, I would not use the word "cause" without a supporting experimental research design (which is a different issue than the question of mediation).
This was so fantastic! Thank you
Thanks for the kind words!
What does it mean when the effect is negative? So the group scoring below the average of the moderator has a negative effect?
Negative predictors are interpreted: as X increases, Y decreases.
Dr. Buchanan, Thank you so much for this video. It has been extremely helpful. I am writing up the results for my moderation model and I was wondering if you had citations for the process in which you screened and removed any leveraging outliers prior to creating running the moderation. Any information would be helpful. thank you.
You can use Cohen Cohen Aiken and West for leverage/cooks and Tabachnick and Fidell for the mahalanobis.
@@StatisticsofDOOM Thank you again!
Great lecture, thanks!
Thank you!
Thanks Erin, great video. I need some clarification on the POWER that you touched upon earlier in the video. As far as predictors (K) are concerned, if I have one IV, one moderator, 1 mediator, and 1 DV, what is the total number of predictors? In other words, sample size required for the medium effect size and power .8. Thanks for your help...
I cannot like this video enough!!
Erin, Wonderful Video. Kindly let me know how to interpret for a continuous independent variable since the video is for a categorical variable.
Hi this video is extremely helpful but I have a question, what is the values in the coefficients column are negative, what do they imply? Thank you so much!
You would interpret that just like a negative correlation ... as X goes up Y goes down.
Thank you so much for this video, which is so helpful to my essay writing.
Glad to be of help!
Hi Erin, when explaining how to report the interaction, you interpret coeff as b (standardized coefficients) while (correct me if I'm wrong) they are unstandardized coefficients in fact.I would like to be able to interpret the strength of the predictor (X) and also see if it is comparable to the previous findings on this topic. If you could relate to this issue, would be great. Thanks!
Right, so in process they are mean centered, and not z-scored. They are unstandardized, but definitely 0 centered.
Thanks for sharing, Erin! Love your video.
Thank you so much Dr. Buchanan, your video is excellent and very very helpful. I have a question that maybe you can help with. I ran moderation and the interaction effect was non-significant, however; in the conditional effect table (high and low groups), all effects become statistically significant. Does this mean I can use this and say moderation does occur for these specific groups?
No - that means you do not have significant differences between slopes, but your main effect of your variable is overall significant.
This was so helpful, thank you! My interaction effect was not significant. Would this be a reason for not seeing the johnson neyman output or the conditional effects? I should add that I am also using the new version3.3
Yes, I think you only get the Johnson Neyman when the interaction is significant. There's a button you can tell it to give you the interaction either way.
Hey Erin! Wonderful video, extremely useful! I do have a quick question though - I have one continuous moderator and one categorical IV with four levels. I am guessing I have to dummy code the variables and then enter those into the model? Could you please throw some light on this? My DVs are continuous too. Thanks!
I have significant X-> Y effect, Johnson-Neyman output, and for simple slopes, significant p-values for high and centered average of my mediator. However, my interaction was not significant. What does this mean, and can I still interpret my simple slopes?
i have similar results. The independent regression coefficients are significant but the interaction is non significant. Can i interpret moderation still? as some books say that moderation is actually the interaction between the IV and Moderator.
That means that there is main effect of X to Y, but the slopes do not different across M. I would not interpret the simple slopes because the interaction is not significant.
I would not interpret the simple slopes if the interaction coefficient is not significant, as that means all the simple slopes are statistically equal (given power, of course) ... the interaction is between X and M predicting Y, correct, which is how it is covered in this video.
Great video! I can't get the "Conditional effect of X on Y at values of the moderator" output in version 3.5 of PROCESS? Can someone help me on how to get this so I can measure the slopes like in the video?
There’s an option to change the p value at which the slopes print out. Otherwise if the moderator is not below that p value it won’t print out.
Thank you Dr Buchanan.
Thanks for a great video. I have a question related to slope. Can I consider the Johnson's values as LOW and HIGH for a slope? Thanks.
1000th like! Well deserved for a very informative lecture (:
Thanks for the kind words!
Hi Erin, I have a question regarding the interaction term in the output. Do I have to include it in my write up and what does it mean if the overall moderation model is significant but the interaction term is not? Thank you very much in advance, Max.
Hi Erin, Thanks for the videos. They are all very helpful! I was wondering whether you were planning to do more videos on PROCESS, especially on how to use it when you have an indip. variable with 3 categories (moderator and dependent variable continuous). Looking in the internet it seems that many people are asking the same question. Plus how do you to report PROCESS results in APA style? Any help would be much appreciate it. Thanks!
Hi Erin!
Thanks for all your videos that helped me a lot! But just one question: Does there actually a graphical user interface for PROCESS exist? Or should I better use SmartPLS or AMOS for this? The problem is that with these programs it is not possible to calculate with moderations.... only mediations are possible (as far as I know)! :((
I'm sorry, I don't understand the question. PROCESS is graphical through SPSS and does both mediation and moderation.
I am sorry for being imprecise, but I meant whether it is possible to "draw" / visualise your calculated model with PROCESS. Like it is in AMOS (you have the regressions of the SEM AND the graphical model with its arrows and boxes).
Have a nice holiday.... ;-)
No on the graphic interface to Process. I think you might be able to coerce AMOS into interactions, but you would have to make the interaction column manually, and then interpreting would be much harder than the regression way (also df is treated very differently).
Also, sorry, I wasn't thinking the right way earlier :| thanks for the clarification!
I really like your video. It helps me a lot. There are still some question remaining though:
1. How do i identify non-significance? Just by looking at the p-value? Or is the LLCI and ULCI also important?
2. What if my overall Effect is non significant, but one of my moderation effet in the slopes is significant? Am i allowed to predict a relationship between my X on Y in that case?
I´d be glad for some help. Thank you!
1) you could do either depending on your area's focus on p values, I would say p values are more comment.
2) that would depend on if the interaction was significant - in general main effects and simple slopes are not interpreted the same.
Thank you Erin, your answer helped me a lot. However when analysing my last Hypothesis i stumbled across another problem:
The moderator is gender. My interaction is non significant (p = .95), but one of my two predictors is significant (p
Hi again, the interaction isn't significant because those two intervals overlap ... so the main effect of that variable is significant (your p
Thanks a lot! You've been of great help! Keep up the good work. Best regards
Hello Dr. Buchanan- I will echo the sentiment of many others here and thank you deeply for sharing this video. It has been crucial in the process of proposing and defending my dissertation.
I have one question- In the initial screen when opening up the process macro, under "options", I have a drop-down menu for "Heteroscedasticity-consistent inference", rather than a checkbox for "Heteroscedasticity-consistent SEs" like in your video. This drop down menu presents several options such as HCO, HC1, HC2, HC3, HC4, as well as "none" (some of these different HC's also have names in parentheses afterwards, I'm assuming credited to different researchers). Do you have any insight about which one of these options I should choose? Thanks!
Let me refer you to Hayes' publication on this topic:
link.springer.com/content/pdf/10.3758/BF03192961.pdf
Specifically I find this paragraph useful:
For small sample sizes, the standard errors from HC0 are quite biased, usually downward, and this results in overly liberal inferences in regression models (see, e.g., Bera, Suprayitno, & Premaratne, 2002; Chesher & Jewitt, 1987; Cribari-Neto, Ferrari, & Cordeiro, 2000; Cribari-Neto & Zarkos, 2001; Furno, 1996). But HC0 is a consistent estimator when the errors are heteroskedastic; that is, the bias shrinks with increasing sample size. Three alternative estimators, HC1, HC2, and HC3, are all asymptotically equivalent to HC0 but have far superior small sample properties relative to HC0 (Long & Ervin, 2000; MacKin-non & White, 1985). A newer estimator, HC4, is preferred when there are cases with high leverage.
Hi. Thank yuou for your video. I have a problem and if you answer ı would be pleased. My p value in model summary is significant however interaction is not significant. So how can ı interpret it? I know there is no mderation but can ı say that x effects y directly from looking at model summary? Thank you.
You will need to see if the main effects are significant to say x predicts y, look at that predictor value.
Very clear delineation. Thank you!
Hi, amazing video - really helped me with many of my projects so far!
Quick question regarding the outlier "analysis": Is there a reference I can use (as in research paper I can quote) that suggests the 2/3 rule for Mahalanobis, Cooks, and Leverage? Personally, I have only found academic articles that suggest to either use only one or maybe two of them. Thanks in advance for any help! :)
I know people have cited my videos, but for academic sources, I usually use Tabachnick and Fidell's book along with Cohen et al.'s book as citations for the outliers. I usually also state that I use 2/3 to adequately cover the different properties of regression (leverage, discrepancy, and patterns for Mahalanobis) without being too sensitive by only using one.
@@StatisticsofDOOM Thank you very much!
Hei Erin
Tank you so much for this video. It was very helpfull. One question tho. When we tried to make the graph and typed in low, average and high i variable view, it didnt show in dataview when we typed in -1, 0 and 1. It this some kind of setting, or do you think we did something wrong?
Hello Dr. Buchanan. Thank you so much for posting this excellent video! It is very clear and easy to follow. I just wanted to make sure I am understanding my results correctly. My overall model was significant, as was my predictor variable (X) and the moderator (M), however, the interaction itself was not significant. Is it accurate to state the predictor variable (X) and the moderator variable (M) predict Y separately? Also, because the interaction is not significant, it is not appropriate to interpret the simple slopes or the J-N output? Thank you in advance for your guidance!
Exactly what you said - x and m predict separately in the main effects, but there is not an interaction. I would not interpret the JN or the simple slope output because they are follow up tests (like ANOVA).
Hi thank you so for this amazing video, just had a question if in multiple regression (step wise) the IV is not a significant predictor can we still use it in moderation analysis and get significant moderation effects?
Yep! Sometimes it only will show an interaction effect with another variable but not a main effect.
@@StatisticsofDOOM ok thank you but then how do we take out effect size then as there are is no significant prediction in regression model?
@@user-iw5xr5ew3f I'm not sure what you mean "take out effect size"?
Hi, thank you for the video, it's great! When calculating a sample size testing more than one moderator, how many predictors do you put in? (I am testing 3 moderators on a linear relationship)
I would figure out how many predictors that would be (a lot with all the interactions), and then use that number.