Hi Dr. Swan, These videos are helpful! I had a question. I am conducting a study that involves a 2x2x2 between-subjects factorial design (so, 3 independent variables), and I was wondering what to run in G*Power to get my needed sample size. I have 4 dependent variables so I'm assuming I need to run a MANOVA, correct? For clarification, is "number of groups" the same thing as number of conditions (i.e. I would enter 8 for that) or is it number of total levels for all of my IVs (i.e. I would enter 6 since each IV has 2 levels)? Also, what does "number of measurements" mean? Is that the number of times a participant is exposed to a condition or something else? Thanks in advance!
Hi Patrick, If you have 4 dependent variables and you're interested in how the composite of the 4 DVs looks linearly, then yes, do a MANOVA. If the DVs are not related to one another, then a MANOVA is not a good idea. Just do 4 separate factorial ANOVAs. Good clarification question, as I went quickly through it: "number of groups" is how many separate groups you have, so yes 8 in your case with a 2x2x2 b/s design. "Number of measurements" is a little stickier, because there's not a lot of good info on the net or from the devs. My ASSUMPTION in this module is that reflects the number of DVs you have in your design, as I said in the video for this part. Essentially, in a fully b/s design, how many times is a person measured? In your case that would be 4. This brings up an important point for anyone watching this video and reading these comments: simpler is better. If you're really interested in a single effect/hypothesis, do the least complex a priori power analysis to accomplish that goal.
Hello professor swan, I have a question concerning the number of dependent variables I am measuring. I want to conduct a repeated measures MANOVA with 4 dependent variables. I only have one group that is measured 2 times. Where do I put the number of dependent variables in G Power when calculating the sample size for the within factor? I can't find a slot... Or is it even necessary to put the number of dv in there? Thank you so much in advance! Best regards, Lisa
Apologies for the delay in my commenting: I would focus on your smallest between-subjects effect, and if there is none, the smallest anticipated effect from one of your within-subjects variables. Any other effect is going to go be generally covered by the power found in the smallest effect. Number of dependent variables are usually put in # of Response Variables, but if you’re using Repeated Measures between factors, the DVs are irrelevant m, so you don’t enter them.
Hello Professor Swan, Im doing a study with 4 groups (2x2) and two related DV's. I do not understand fully what is meant bij Response Variables? My squared effected size would be large (0.67). Is it also possible to just use the standard squared effect size given?
Response variables are your dependent variables. Participants "respond" to these variables. You are welcome to use conventions for your power analyses, as long as you try your best to stick to the plan!
Hello professor Swan, Thank you so much for this clear video. I have a question about the response variables, because how do you come up with 2 response variables? In my case, I have 3 questionnaires: the first questionnaire has 4 subscales, the second questionnaire has 2, the third questionnaire has 2. So how many response variables do I have? 8 in total? And do you also need to see the covariates as response variables? Thank you in advance! Best wishes, Manon
I used "response" variable to mean dependent variable in this context. The question you need to ask yourself regarding # of variables is whether these subscales are collapsed into composite scores. For example if your first questionnaire hs 4 subscales, but those subscales are averaged or summed into a final composite score to be analyzed, you have only 1 response variable for that particular questionnaire. From what I can see, at minimum, you have 3, at most you have 8. It depends on what you do with these subscales in the end.
Hello Professor, Thanks for the tutorial! I have a question regarding which type of statistical test to choose in G*power. I have two DVs (gender [3 levels], academic year [4 levels]) and two IVs, one IV has 4 levels while the other has 2 levels. I want to check for gender differences and year-wise differences in my two IVs, so will go for two-way MANOVA (within group). In G*power, would I need to go for global or special effects? Also, what would the number of groups and response variables be? Any help or advice is appreciated! Best wishes!
Gender shouldn't be a DV; though it is a measured variable, it is an existing quality of a participant, and therefore can't be an effect (typically what a DV is). You want Gender to be one of your IVs, which would lead you to have 3 IVs and one DV. So a factorial ANOVA. But even year diff really shouldn't be a DV either. You may want to reconsider your design from the info you've shared with me. Send me an email or find me on my Discord if you want to discuss further.
@@AlexanderSwan hi, thank you for your reply, gender is actually my IV, I made an error typing this out. As for academic year, it was also an IV but I have removed it from my plan. So, only gender as an IV will be tested against two DV's which each have 4 subscales.
Hello! Thank you so much for this video. Could you provide some assistance in conducting the post-hoc analysis for the MANOVA global effects test as I am looking to calculate power? My output from SPSS reported my effect size as partial eta squared (i.e., η^2) rather than the effect size of f^2(V), so I got confused. I have 2 levels for my IV and 4 dependent variables. My TOTAL sample size was 40. Do I look at my MANOVA output and take the number of Pillai's Trace and enter that to calculate my effect size in terms of f^2? Any help is appreciated. Thank you!
I got confused when you talked about pillai v in manova between factor. I thought it was different, and that was the cause of a high Number of individuals (90), instead of 21 as erlier.
Hello Professor Swan, I am a student and want to conduct an experiment with one independent variable, one moderator (nominal scale), three dependent variables and several covariates. The problem is that the moderator only interacts with one level of the independent variable, so I have an asymmetrical design with only three groups (instead of the usual 2x2). Can I still use MANCOVA and can I use the same inputs in G Power as with a usual factorial design?
Does the moderating variable exist in a present/absent kind of way? The way you've described it sounds confounded. I'd be happy to help but I think I need more details before providing any advice. Email me or or catch me on my discord under stats-help. Invite link in description.
Hello professor Swan, G* Power is excellent for calculating required sample size for hotelling T^2: two group mean vectors for me. Thanks a lot. However, I need to consider different effect size in my studies because I want to calculate different required sample size for different experiment. Is there a way for me to use G* Power in Python ? Best regards, Xin-Ping Chen studying electronics engineering in Taiwan
Hello professor Swan, it's me again I wonder whether the "effect size" in MANOVA Power Analysis equals the Mahalanobis distance between two population. Here is the definition of Mahalanobis distance en.wikipedia.org/wiki/Mahalanobis_distance . From my own experiments, I found that G* Power often underestimates Power of Hotelling T^2, so I am not sure whether I mistake effect size as Mahalanobis distance. Best regards, Xin-Ping Chen
Hi there, I'm afraid I don't know the answer to your question. I've never used MHD as an effect size, and do not do MANOVAs in my research (there are a number of issues with the technique, for which HLM is better suited). The devs of GPower may know the answer as to what they programmed for effect size better than I!
Hi there, i have a question regarding which statistical test to use. I have one independent variable which is condition (3 conditions), I also have 3 separate dependent variables. Which MANOVA option would it be on GPower and what would the specific input parameters be? i am uncertain about whether the default inputs are correct. Any help would be greatly appreciated, thank you :)
Hmm, you used "separate" DVs... are you interested in the linear combo of these variables? That's the only reason to use a MANOVA. Otherwise I would suggest three separate ANOVAs. Also, is your IV between- or within-subjects?
hey there, I have a question, I have to conduct 2 way manova, I have 2 IVs and 4 DVs, in one IV there will be control and experimental group, my second IV is gender and Im going to look for that in both control and experimental groups. I thought that my analysis is going to be special effects and interactions but I couldnt be sure. I will be glad if you help me out!
Go with the power analysis of the most important effect you're looking for or the smallest effect you expect. So if the interaction of gender and condition is what you're after, go for that.
Hi Dr. Swan,
These videos are helpful! I had a question.
I am conducting a study that involves a 2x2x2 between-subjects factorial design (so, 3 independent variables), and I was wondering what to run in G*Power to get my needed sample size. I have 4 dependent variables so I'm assuming I need to run a MANOVA, correct?
For clarification, is "number of groups" the same thing as number of conditions (i.e. I would enter 8 for that) or is it number of total levels for all of my IVs (i.e. I would enter 6 since each IV has 2 levels)? Also, what does "number of measurements" mean? Is that the number of times a participant is exposed to a condition or something else?
Thanks in advance!
Hi Patrick,
If you have 4 dependent variables and you're interested in how the composite of the 4 DVs looks linearly, then yes, do a MANOVA. If the DVs are not related to one another, then a MANOVA is not a good idea. Just do 4 separate factorial ANOVAs.
Good clarification question, as I went quickly through it: "number of groups" is how many separate groups you have, so yes 8 in your case with a 2x2x2 b/s design. "Number of measurements" is a little stickier, because there's not a lot of good info on the net or from the devs. My ASSUMPTION in this module is that reflects the number of DVs you have in your design, as I said in the video for this part. Essentially, in a fully b/s design, how many times is a person measured? In your case that would be 4.
This brings up an important point for anyone watching this video and reading these comments: simpler is better. If you're really interested in a single effect/hypothesis, do the least complex a priori power analysis to accomplish that goal.
@@AlexanderSwan Perfect! Thank you for your help, Dr. Swan!
Hello professor swan,
I have a question concerning the number of dependent variables I am measuring.
I want to conduct a repeated measures MANOVA with 4 dependent variables. I only have one group that is measured 2 times. Where do I put the number of dependent variables in G Power when calculating the sample size for the within factor? I can't find a slot... Or is it even necessary to put the number of dv in there?
Thank you so much in advance!
Best regards, Lisa
Apologies for the delay in my commenting: I would focus on your smallest between-subjects effect, and if there is none, the smallest anticipated effect from one of your within-subjects variables. Any other effect is going to go be generally covered by the power found in the smallest effect.
Number of dependent variables are usually put in # of Response Variables, but if you’re using Repeated Measures between factors, the DVs are irrelevant m, so you don’t enter them.
Hello Professor Swan,
Im doing a study with 4 groups (2x2) and two related DV's. I do not understand fully what is meant bij Response Variables? My squared effected size would be large (0.67). Is it also possible to just use the standard squared effect size given?
Response variables are your dependent variables. Participants "respond" to these variables. You are welcome to use conventions for your power analyses, as long as you try your best to stick to the plan!
Hello professor Swan,
Thank you so much for this clear video. I have a question about the response variables, because how do you come up with 2 response variables? In my case, I have 3 questionnaires: the first questionnaire has 4 subscales, the second questionnaire has 2, the third questionnaire has 2. So how many response variables do I have? 8 in total? And do you also need to see the covariates as response variables? Thank you in advance!
Best wishes, Manon
I used "response" variable to mean dependent variable in this context. The question you need to ask yourself regarding # of variables is whether these subscales are collapsed into composite scores. For example if your first questionnaire hs 4 subscales, but those subscales are averaged or summed into a final composite score to be analyzed, you have only 1 response variable for that particular questionnaire.
From what I can see, at minimum, you have 3, at most you have 8. It depends on what you do with these subscales in the end.
Hello Professor,
Thanks for the tutorial! I have a question regarding which type of statistical test to choose in G*power. I have two DVs (gender [3 levels], academic year [4 levels]) and two IVs, one IV has 4 levels while the other has 2 levels. I want to check for gender differences and year-wise differences in my two IVs, so will go for two-way MANOVA (within group). In G*power, would I need to go for global or special effects? Also, what would the number of groups and response variables be?
Any help or advice is appreciated!
Best wishes!
Gender shouldn't be a DV; though it is a measured variable, it is an existing quality of a participant, and therefore can't be an effect (typically what a DV is). You want Gender to be one of your IVs, which would lead you to have 3 IVs and one DV. So a factorial ANOVA. But even year diff really shouldn't be a DV either. You may want to reconsider your design from the info you've shared with me. Send me an email or find me on my Discord if you want to discuss further.
@@AlexanderSwan hi, thank you for your reply, gender is actually my IV, I made an error typing this out. As for academic year, it was also an IV but I have removed it from my plan. So, only gender as an IV will be tested against two DV's which each have 4 subscales.
Hello!
Thank you so much for this video. Could you provide some assistance in conducting the post-hoc analysis for the MANOVA global effects test as I am looking to calculate power?
My output from SPSS reported my effect size as partial eta squared (i.e., η^2) rather than the effect size of f^2(V), so I got confused. I have 2 levels for my IV and 4 dependent variables. My TOTAL sample size was 40. Do I look at my MANOVA output and take the number of Pillai's Trace and enter that to calculate my effect size in terms of f^2? Any help is appreciated. Thank you!
I’m not sure I’m following your question here. What do you need to calculate exactly and for what PA?
I got confused when you talked about pillai v in manova between factor. I thought it was different, and that was the cause of a high Number of individuals (90), instead of 21 as erlier.
I don't know if I understand your question/confusion, but B/S designs likely always need more people than a w/s design.
Hello Professor Swan,
I am a student and want to conduct an experiment with one independent variable, one moderator (nominal scale), three dependent variables and several covariates.
The problem is that the moderator only interacts with one level of the independent variable, so I have an asymmetrical design with only three groups (instead of the usual 2x2).
Can I still use MANCOVA and can I use the same inputs in G Power as with a usual factorial design?
Does the moderating variable exist in a present/absent kind of way? The way you've described it sounds confounded. I'd be happy to help but I think I need more details before providing any advice. Email me or or catch me on my discord under stats-help. Invite link in description.
Hello professor Swan,
G* Power is excellent for calculating required sample size for hotelling T^2: two group mean vectors for me. Thanks a lot. However, I need to consider different effect size in my studies because I want to calculate different required sample size for different experiment. Is there a way for me to use G* Power in Python ?
Best regards,
Xin-Ping Chen studying electronics engineering in Taiwan
I don't know if G*Power connects to Python, I'm not a Python user. Good luck!
Hello professor Swan, it's me again
I wonder whether the "effect size" in MANOVA Power Analysis equals the Mahalanobis distance between two population. Here is the definition of Mahalanobis distance en.wikipedia.org/wiki/Mahalanobis_distance . From my own experiments, I found that G* Power often underestimates Power of Hotelling T^2, so I am not sure whether I mistake effect size as Mahalanobis distance.
Best regards,
Xin-Ping Chen
Hi there, I'm afraid I don't know the answer to your question. I've never used MHD as an effect size, and do not do MANOVAs in my research (there are a number of issues with the technique, for which HLM is better suited). The devs of GPower may know the answer as to what they programmed for effect size better than I!
@@AlexanderSwan Thank you so much. Does HLM mean hierarchical linear modeling?
@@user-ch6gk5gw7k Oh yes, that's what I meant by HLM.
Hi there, i have a question regarding which statistical test to use. I have one independent variable which is condition (3 conditions), I also have 3 separate dependent variables. Which MANOVA option would it be on GPower and what would the specific input parameters be? i am uncertain about whether the default inputs are correct.
Any help would be greatly appreciated, thank you :)
Hmm, you used "separate" DVs... are you interested in the linear combo of these variables? That's the only reason to use a MANOVA. Otherwise I would suggest three separate ANOVAs. Also, is your IV between- or within-subjects?
hey there, I have a question, I have to conduct 2 way manova, I have 2 IVs and 4 DVs, in one IV there will be control and experimental group, my second IV is gender and Im going to look for that in both control and experimental groups. I thought that my analysis is going to be special effects and interactions but I couldnt be sure. I will be glad if you help me out!
Go with the power analysis of the most important effect you're looking for or the smallest effect you expect. So if the interaction of gender and condition is what you're after, go for that.