SmartPLS 3 Multigroup Analysis MGA

[@Gaskination]

Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the KZbin channel where we post a new video almost three times per week: kzbin.info/door/iujxblFduQz8V4xHjMzyzQ Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074 And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en Check it out! Thanks!

[@moayad2055]

man, you are really nice guy. helped me a lot. God bless you.

[@Florian_Sieben]

Thank you so much! Perfect tutorial . This helped me a lot. You’re a good teacher 👍

[@samrach1120]

Dear Dr James. I need to compare three groups. if possible can you upload a video on omnibus test or refer me to a video where i can learn that from ? I couldn't find any video on that.

[@chefberrypassionateresearcher]

Professor, how to compare groups if i have 3 groups?

[@Gaskination]

Do it in pairs. Or, use the new feature in SmartPLS 4 that allows you to compare all groups at once. I don't think I have a video for that yet. Just check all the group boxes when setting up the MGA.

[@chefberrypassionateresearcher]

@@Gaskination Prof, thanks for your able guidance. Please tell if i want to do Kruskal wallis test, shall i do MICOM(measurement invariance first, so that any insignificant items may be removed from the dataset during the second step of MICOM? After this i may go in for ANOVA, or descriptives. Is that right?

[@Gaskination]

@@chefberrypassionateresearcher I'm not sure. I never use the Kruskal Wallis test. Here's what I learned about it just now: Kruskal-Wallis Test The Kruskal-Wallis test is a non-parametric method for testing whether samples originate from the same distribution. It is used when the assumptions of one-way ANOVA are not met, particularly when the data is not normally distributed. Purpose: To compare three or more independent samples to determine if they come from the same distribution. How it works: It ranks all the data across the groups together and then checks for differences between the sums of these ranks for each group. Significant differences between these sums suggest differing distributions. Usage: It’s useful when data are ordinal or when the scale of measurement does not meet the assumptions of parametric tests.

[@사브리나-u5u]

Hi. I want to ask you. I noticed that one hypo wasn't significant in both males , and females. But it showed a significant number if I merge data. Does it make sense? I worried about it?

[@Gaskination]

That is probalby due to sample size and error. The smaller samples (when split) have just enough error to reduce the critical value (and hence raise the p-value).

[@HeroHero64]

Dear James Gaskin, thank you for your videos. I have a question. In the case of a 2 x 2 between-subjects experimental design, is it possible to analyze the interaction effects of these two factors on DVs using MGA?

[@Gaskination]

Yes, that should work.

[@leovaniguimaraes3491]

Hello dear Dr. Gaskin! I am facing a singular matrix issue when running PLS-MGA in a model with one formative exogenous composite with 11 indicators. I have 2 groups, 235 observations in each, The problem is that I tried the obvious things to fix that, like indicator with no variance, etc. but I can not fix it. And the error only occurs if I use the formative measurement for this exogenous variable. If I invert the model, the problem disappears. Please, could you give me some help?

[@Gaskination]

That's weird. A singularity matrix usually occurs when a sample size is insufficiently small to support too much error in the data. This can also be a result of a variable that does vary for one of the groups (e.g., if you included the grouping variable in the model as a factor).

[@samrach1120]

Also (I could be wrong) I tried MGA in PLS with three levels (high, medium and low). In both groups A and B , check all three levels and run MGA. It does give the difference of path coefficients ( High vs Low, High vs Med, and Low vs Med) , and the p values of the differences ( High vs Low, High vs Med, and Low vs Med). I assume this is doable like this.

[@kanwalnasim9195]

Can anyone please tell me which data file do we have to use here? Data generated by PLS (that we used in path model analysis) or initial data?

[@Gaskination]

MGA can be done with your initial data. The only reason you would instead use data generated by PLS is if you needed to do a two stage approach due to a formative higher order factor being predicted by other factors.

[@kanwalnasim9195]

@@Gaskination thanks for your reply. I have second and third order constructs. For my base model I used data generated by PLS, as shown in your video. Now I want to run MGA on same model. So, with data should I use now?

[@Gaskination]

@@kanwalnasim9195 Use the new data (with latent variable scores)

[@marynn89]

Dear Dr James, I have a model that consists of two IV's ('Benefits' and 'Risks'), one DV (Willingness to share information) and one moderator('Type of data'). The Benefits (IV) variable is categorical, where 0 = Non material benefit and 1 = Material benefit. Now I want to test if the different types of Benefits have a positive and significant correlation with the DV. If i run the PLS Algorithm and Bootstrap I only see one path coefficient (0.222) and T value (3.4). How do I know if this are the values for Material (1) or Non Material benefit (0)? And how can I find the values for the other type of benefit? My moderator variable is also categorical (Low sensitive = 0 vs. High sensitive =1). I want to test if high sensitive information has a negative impact on the Material benefit > Willingness to share and Non material benefit > willingness to share effects, should I use multi group analysis in this case?

[@Gaskination]

When an IV is nominal (two unordered categories), it may be better to shift it to the moderator position. In this case, you would have groups representing the following: -Material, low sensitive -Material, high sensitive -Nonmaterial, low sensitive -Nonmaterial, high sensitive As long as you have enough sample size to cover these groups, this will make it easier for you to test your hypotheses.

[@jinalsameershah]

I was applying Micom in smart pls and in step 2 for checking compositional in variance.. I am getting original correlation as 1. Is this correct output. Or there is an issue. Can I still go ahead with MGA

[@Gaskination]

Correlations should only be 1 if it is a self-correlation.

[@psrivastava9893]

Hi Prof., Just one quick question, the bootstrapping result for second and third provided that the path is not significant for high frequency and low frequency group. So, if we have found that the path of one group or both the group is not sig., do we still check the parametric result or it is of no use if any one group is not sig. ?

[@Gaskination]

You can check it and say that the IV is a good predictor for only one group. This is a weaker form of moderation.

[@gingerginger733]

Thank you Professor. I wonder how to do MGA in high order contruct model.

[@Gaskination]

Same way as for the model in this video above. The only complication is if you have a higher order formative factor that is also being predicted by other factors. In such a case, you would need to take a two step approach as shown in this video: kzbin.info/www/bejne/goOxdWB-Ys2Eh9k and then do the MGA on the LVS model (in the second stage).

[@muhammadzeshan2727]

Hi, I hope you are doing great. I want to compare the SEM results for two countries. Can I do that in AMOS?

[@Gaskination]

Yes. Here is a video: kzbin.info/www/bejne/rWbMnKJ_nLl8mZI

[@parveenaasalam7040]

Can we accept hypotheses at less than 0.1 level?

[@Gaskination]

If you mean a p-value less than 0.10, then yes, if it makes sense for your context.

[@parveenaasalam7040]

@@Gaskination Thank you very much

[@amalbakry5974]

Dear Sir in my model i have six independent variables and three moderators and one dependent variable , all variables are latent variables so for moderation analysis can i use subgroup or run interaction moderation as i will have huge numbers of interaction moderation.

[@Gaskination]

If using interactions for moderation, you can run these one at a time (exclude other moderators). If using MGA, yes, make sure to run these for one moderator at a time.

[@amalbakry5974]

Dear Sir , i have two questions : 1- should i include the latent reflective moderator variable in measurement model analysis like convergent validity and discriminat validity. 2- if i have converted the latent reflective moderators to composite one variable and split the data by median can i do the invariance analysis for multigroup for every moderator if i have more than one moderators. thanks a lot for usual support

[@Gaskination]

@@amalbakry5974 1. The "correct" answer is yes, it should be included and validated. The practical answer is that it is optional if you are just going to collapse it and split it. 2. Yes, you can do the invariance test for every moderator, but I would recommend doing it with just two groups at a time (it's just easier to understand this way)

[@amalbakry5974]

@@Gaskination thank you so much for reply but I have another clarification that when doing CFA in smartpls you draw the model so for moderators variable can I draw an arrow from moderators to dependent variable only or must define it as a moderators to do convergent validty and discriminat validty and thanks again for your usual support and God bless you 🙏

[@Gaskination]

@@amalbakry5974 If it is a latent factor, then it would be good to include it when testing validities.

[@irnalestari6126]

Hi Prof, Thanks for the video. I use Number of Kids as moderating variable, and it generate into 4 groups. I can do data group bootstrapping for all this 4 group. But when i tried to run MGA, there is no option to choose more than 2 groups. So how should i do the hypothesis test for each group of moderating variable? Many thanks in advance.

[@Gaskination]

When there are more than two groups, you have to do it in pairs. So, e.g., A:B, A:C, A:D, B:C, B:D, C:D. Or do one vs other; e.g., A:BCD, B:ACD, C:ABD, D:ABC.

[@irnalestari6126]

@@Gaskination Thanks a lot Prof

[@MB-ui7er]

@@Gaskination Any reference video on this for three groups as a moderator? I have moderator(short term , mid-term and long term investment. Please if there is any tutorial on this share it with me.

[@Gaskination]

@@MB-ui7er For more than two groups, just compare them in pairs.

[@parveenaasalam7040]

Can comparative analysis be done by multigrouo analysis on smart pls?

[@Gaskination]

Yes, we can compare two contexts or groups by using MGA in SmartPLS.

[@parveenaasalam7040]

@@Gaskination Thank you very much

[@monaisazad34]

​ @James Gaskin Dear James thanks for the great video. I have some questions regarding a second-order reflective- reflective (all second-order constructs are exogenous) model. 1) If we want to check the moderation effect of gender on such a model, do we still need to go through the latent variable scores? Can you please guide me in this regard? 2) And how to interpret the results? 3) Also, in MGA, do we need to check the confidence interval or only the p-value is enough? Thanks in advance for all your effort and for sharing your knowledge with us.

[@Gaskination]

1. No. You only need LVS when the endogenous variable is formative 2nd order. 2. Results can be interpreted just as they are in this video. No special considerations need to be made when it is reflective-reflective. 3. Confidence intervals are always informative and useful.

[@monaisazad34]

@@Gaskination Dear James Thank you so much for your kind and detailed response. I really appreciate that.

[@ramazanhilmi]

Hi Prof., how shall we proceed to do MGA with second order factor with latent variable scores? Can we add group variables to the latent scores raw file? or is there another way to do it?

[@Gaskination]

If you mean your moderator is a 2nd order factor, then don't do multigroup with it. Do interaction instead. If you mean the other variables are 2nd order, then you can run the model as usual, get to stage 2 (with the latent variable scores) and then do MGA.

[@ramazanhilmi]

@@Gaskination Thank you a lot for your reply. I am doing it with gender. I added gender variable to the raw file of latent scores that was calculated for the 2nd stage, assuming the order of latent scores are the same with my original raw data. The MGA works now. Is this correct way to do it?

[@Gaskination]

@@ramazanhilmi Yes, they are the same order. That is correct.

[@tasneemsaifuddin5862]

I have to analyze the effect of assessments on primary and secondary institutions means a compartive study on primary and secondary institutions , how can I analyze with the help of smart pls

[@Gaskination]

If you want to see how assessment performance differs across institutions, then just use an ANOVA. If you want to see how variables are related, then you can do a path analysis or latent causal model in SmartPLS and use the MGA approach shown in this video.

[@tasneemsaifuddin5862]

@@Gaskination thank you so much for your reply

[@PraveenSrivastavaBITMesra]

(At 5:00 min.), while configuring the data group, I think, if we are having Less than 5 for one group, it should be more than 4 for the second rather than more than 6.00. Then we can get the value for total cases.

[@MOHAMEDAMINE-dw8lt]

Hello Dr. Gaskin. I am doing a SEM modeling using SamrtPLS. In my reserach I have categorized the Company size group into three groups such as large, Medium, small I want to test my model with Company size group (large, Medium, small). . How can I test the model with difference of Company size group among three categories.

[@Gaskination]

MOHAMED Alili unfortunately, you must compare only two things at a time. That’s how the human mind works anyway though. We can only compare two things in our head at once. It is the same with the algorithm. So, compare small to big, and then big to medium, and then small to medium.

[@MOHAMEDAMINE-dw8lt]

thank you dr

[@monaisazad34]

Thanks for the video. I have a question. What if we want to check the moderation effect for numeric variable or for categorical variables with more than 2 levels. how should we interpret the results? For example, if for two age groups 20-30 and 30-40 there is a significant difference, but between 30-40 and 40-50 there is not; can I argue that age has a moderation effect; Or I just can say for that two specific groups (20-30 and 30-40) there is moderation effect of age? Thanks in advance for your help.

[@Gaskination]

As you note, you'll need to compare them in pairs. SmartPLS (and no other software that I know of) does multi-way comparisons. As for the reporting, more information is better. So, yes, report that age is a moderator, but only for certain pairs.

[@monaisazad34]

​@@Gaskinationthanks

[@samrach1120]

One last question , can we make graphs of multigroup analysis (lets say for two groups) in PLS ?

[@PraveenSrivastavaBITMesra]

If I need to check the moderating role of Age (6 different age group) in one relation, can I use MGA in Smart PLS ?

[@Gaskination]

You can do it that way with all six groups (but only two at a time). SmartPLS lets you select multiple subgroup to form a single comparison group. So, for example, for group A you might select age groups 1-3 (younger) and for group B you might select age groups 4-6 (older).

[@PraveenSrivastavaBITMesra]

@@Gaskination Thanks again for the timely assistance.

[@educationalconsultant9880]

I am doing comparative analysis on impact of accredited and non accredited teacher education programs on student's achievement. In that case my accreditation variable on the basis of which I am.analysing the comparative result is secondary data.rest all are primary data. Can I run multigroup analysis on this model?

[@Gaskination]

Use multigroup if you want to see if a relationship between variables depends on accreditation. Use ANOVA or t-test if you want to see if the distribution or average levels of variables differ based on accreditation.

[@educationalconsultant9880]

@@Gaskination thank u for ur reply

[@md.mahiuddinsabbir6125]

Dear James, Thank you very much for your tutorials. I have a few questions. I am intending to do MGA with formative 2nd order constructs (Utilitarian and Hedonic quality) as IVs, Perceived usefulness (reflective), Perceived ease of use(reflective), Perceived enjoyment (reflective) as DVs, Product category (Utilitarian and Hedonic) as a moderator. Should I proceed as you described in the current video? Also, if I run MGA with a new set of data then do I need to be concerned about criteria relating to measurement and structural model based on this new data?

[@Gaskination]

Yes, you can use MGA as in this video, and yes, you need to make sure the measurement properties are valid, such as discriminant validity.

[@md.mahiuddinsabbir6125]

@@Gaskination Many thanks, Sir.

[@mortizmortiz]

Good evening. I have one question about MGA analysis including 2nd order constructs. As for the data, did you take the latent variables values ?

[@feliperodriguescruz1794]

Hello Dr. Gaskin. Could you please help me? I need to know how do i deal with Categorical Variables (variables divided in groups) as moderators. I'm using smart PLS 3. Thank you!

[@Gaskination]

That's what this video is all about. Did you a question that was not answered by this video?

[@feliperodriguescruz1794]

Oh, is MGA used as moderation for Categorical Variables ? I'm just starting to learn about PLS and SmartPLS to be honest. Thank you! I'll watch it again more carefully.

[@Gaskination]

Yes. MGA is multi-group analysis, this is just a fancy way of saying categorical moderation.

[@feliperodriguescruz1794]

Thank you alot for all the help (posting videos and replying to questions).. You're the man!

[@hwangyunseong8579]

Hi Dr James, Thank you for a good video! I want to ask, since i use UTAUT2 and they have a moderate variable like age, gender, and experience, how I put that three in the questionnaire so I can processing the data to get the results? Can I only ask them to choose between female or male, or I have to make the answer be 1 or 2 with an explanation 1 for male and 2 for female? Thank you so much hope you answer this!

[@Gaskination]

The current common convention with gender reporting is male, female, nonbinary, other, prefer not to disclose. Choose one only. Age and experience are best asked with a slider (rather than text entry).

[@daydreamingyou]

when I tried to run MGA, I got singular matrix problem, can someone please help me.. I have 400 data, so I think it's enough and don't have to add more, I'm want to try run correlation matrix to solve the problem but idk how to do it

[@Gaskination]

This happens when you include the grouping variable as a predictor in your model and use it in the MGA. For example, if you are using gender as a grouping variable, then gender should not show up as a variable in your model.

[@daydreamingyou]

@@Gaskination omg, I just delete it from my model and it works! so, I just like duplicate my model and delete my moderator and when I run it, it works. So, I have 3 moderator variable, and I thought that for hypothesis analysis I have to stick all variable in one model and run the bootstrapping. So for bootstrapping we just have to run it without moderator and then analysis the moderator using MGA. am I right?

[@Gaskination]

@@daydreamingyou correct. Glad it worked

[@daydreamingyou]

@@Gaskination thank you so much for your help!

[@haohvtc5984]

Thank you so much for your video. Could you please help me this question? I use 15 indicators (7 likert scale) to measure the mindfulness of customers. I want to classify them into two groups: high mindful and low mindful and compare them each other. Could you please suggest me how to classify them?

[@Gaskination]

You could sum or average their responses so that the sum or average represents the extent of mindfulness, and then split them along the median (or along a theoretically chosen middle point).

[@haohvtc5984]

@@Gaskination Thank you so much for your quick reply. I am a novice of PLS but splashing with it. During your videos, you sometimes said it was a pain. Honestly, I have had lots of pains. :(

[@haohvtc5984]

@@Gaskination Could you please help me more details that to split customers into two groups, what should I use - mean or median? I have had findings of path coefficients, and now want to compare the path coefficients between the two groups, so I use Multigroup analysis, right? Lots of pains during my research! Thank you so much!

[@ahmedlouay]

Hi James, Does SmartPLS 3 support ordinal dependent variable? Many thanks in advance.

[@Gaskination]

Yep. All of the variable in this video are ordinal. I think they are on a 5-point Likert scale of agreement.

[@fazlidarazali4687]

Hi Dr, do i still need to generate new data set for men and female if im using PLS 3.0. One more things, do you have a video on how to do moderated mediation MGA for PLS 3.0 really need your guide Dr.

[@Gaskination]

No. As I show in this video, you can just identify those groups in a single dataset. I don't have a video on moderated mediation yet. But you should be able to run MGA on a mediation model. You won't have a statistical test for the difference in the indirect effects across groups, but you will be able to compare those effects for amplitude, direction, and significance.

[@shinajvs527]

Dear Scholars, Is it required to conduct omnibus test of group differences (OTG) if the study compare different categories under one group. For instance, if the study attempts to examine the heterogeneity effect in a single group, namely 'Authority type', and that includes three categories i.e. 1)Government; 2) Semi-government; and 3) Private. Sarstedt, Henseler, and Ringle (2011) proposed the omnibus test of group differences (OTG) for analyzing the differences of two and more groups of relationships. Based on the case mentioned above, according to Sarstedt, Henseler, and Ringle (2011), what is considered as a group? is it 'Authority type' ? or three classication stated under the group? Can i just conduct PLS- MGA in above case?

[@Gaskination]

Regular MGA should be fine. I think the concern is if the groups are not sufficiently distinct at the structural level. When this is the case, there might only be small differences.

[@shinajvs527]

@@Gaskination I do appreciate your prompt response. For further clarification, When do a researcher use omnibus test of group differences (OTG), and what are the advantages over the PLS MGA?

[@Gaskination]

@@shinajvs527 I'm not sure. I've never done an OTG test.

[@shinajvs527]

@@Gaskination Alright..Thank a lot for the prompt response

[@UAEHolden]

Can I rely on PLS MGA tab, or I have to use the Parametric test tab? to answer my research question about gender?

[@Gaskination]

Nasser Almuraqab MGA should be fine.

[@UAEHolden]

James Gaskin thanks, appreciated

[@UAEHolden]

Appreciated. Another question: What about significance level of MGA I have seen another video it is saying if it is %5 < or >95% can you explain me the 95% part, I didnt get it. Thanks in advance

[@Gaskination]

If the p-value is less than 0.05 (i.e., 95% confidence level), then the effect is significant.

[@mastegas]

No measurement invariance test (MICOM) before MGA?

[@Gaskination]

That would be a good idea. I'll have to learn about it and then make a video for it. So far, I feel like a bit of a novice with SmartPLS...

[@shwetajaiswal5781]

Hi Prof, could you please demonstrate or explain how can we analyse data incorporating control variables as well in smart pls

[@TheAISChannel]

control variables can be included just as any independent variable.

[@ManoBurtai]

Dear James Gaskin, thank you for your videos on SmartPLS, they are very helpful! I have one question regarding multigroup analysis. I would like to compare my model based on a few sociodemographic characteristics as well as consumption data that I collected through my survey. As an example, I have 65 men in my sample and 148 women. In regards to 10-times rule, the busiest construct in my model has 27 arrows pointing to it, which implies that I need 270 observations for every group. I am wondering if there is still a way to perform multigroup analysis without meeting this sampling requirement, or it is better to forgo as it will not provide any meaningful results? I would appreciate your expert advice! Thank you so much!

[@Gaskination]

Wow! That's a busy construct... Perhaps it would be best to simplify a bit. Usually when a factor has that many indicators, it implies there are multiple dimensions to it and they should be separated into those dimensions. As for whether you should run it as is, I think you will find a sample size of 65 insufficient to run such a complex model.

[@ManoBurtai]

Dear prof. James Gaskin, thank you for your prompt reply and your insights. It is indeed a very busy construct, and I cannot reduce the number of arrows, unfortunately. I have 27 separate product attributes each being measured with a single-item that are connected to endogenous variables (food consumption goals), and this study (my master thesis) is rather exploratory. Do I understand correctly that I should have at least 200 observations for each group to run multigroup analysis in my case or, for example, 100 observations for each group would also suffice in the worst case scenario? I am still questioning what an appropriate sample size would be for group analysis taken into consideration model complexity, and I cannot find a precise answer in the literature on PLS-SEM. Thank you once again!

[@Gaskination]

There are several way to calculate sample size requirements. The reason you won't find a solid answer in the literature is because it depends a lot on error. If there is very little error in the measures and the relationships, then less sample size is needed. You could probably get away with 100 per group. However, with lots of error, even 200 per group might not be enough.

[@ManoBurtai]

Dear James, thank you so much for your answer!

[@saraosama6550]

Thank you so much, Professor James for the valuable videos. Do you think we can use MGA for analysing between subjects design research. For example, if I have three groups each has been exposed to a different stimulus. Thank you so much in advance.

[@Gaskination]

Probably ANOVA is best for that design. MGA is appropriate if you have a path model you want to compare across groups.

[@saraosama6550]

@@Gaskination Thanks you so much.

[@saraosama6550]

In fact, Professor Gaskin, I do want to compare a path model across the three groups which are exposed to the three conditions.

[@Gaskination]

@@saraosama6550 Then yes, MGA is the right approach

[@saraosama6550]

Thanks a million, Professor.

[@FombapKing]

Thank you, Mr James

[@TankAqil]

Hello, James. I am currently using PLS MGA to see the moderating effect of gender in my dissertation. I have a quick question. If the p-value in "Parametric Test" table is significant, do we interpret that there is a moderating effect? I would greatly appreciate your help. Thank you :)

[@Gaskination]

Yes, look at the parametric test tab. If the p-value is less than 0.05 (or 0.1 if 90% confidence), then the moderation is significant. As for the strength of the moderation, instead, I recommend looking at the strength of the effects between male and female. If the effect is stronger for male, then report that. If stronger for female, then report that.

[@mrli3666]

@@Gaskination hello Prof, have a question.. i use age as moderation effect.. what if the boothstrapping result is for the young category is significant but for the old category is not significant.. and the parametric test result is not significant.. can i interpret that age have no moderating effect? Or i should said only old category have moderating effect but the young category is not significant?

[@Gaskination]

@@mrli3666 This is a form of moderation. It shows that your IV is a good predictor of your DV only for the young group, but not for the old group.

[@mrli3666]

@@Gaskination thankyou for your answer prof! Its really help me 😁

[@jinalsameershah]

Sir, can you make a video on FIMIX ?

[@Gaskination]

If I did, it probably wouldn't be a very good one :) I have never used it. If I do use it in the future, I will make a video for it.

[@dr.mohammadmasudurrahman7401]

Hi Dr. Thank you so much for your video. Recently, I was trying to do moderating by age, experience, and gender. I did MGA for Age and Experience but can't do MGA for Gender, it shows singular matrix problem. Is there any other way to solve this moderating issue for gender?

[@Gaskination]

Make sure gender is not in the model when it is also being used as a grouping variable. Otherwise you end up with a singularity matrix because gender will be all male for the male group (etc.).

[@dr.mohammadmasudurrahman7401]

@@Gaskination Thank you for your reply. I have male student, male faculty, female student, and female faculty as a respondent in my research, so I solved this issue such as male student=1, male faculty=2, female student=3, and female faculty=4 then under gender I made group (male (less than 3) and female (higher than 2)). But if you make a video where we have only male and female categories such as male student and female student only. It will be highly appreciated. Thanks again.

[@jiangpan7333]

Dear Dr. Gaskin For MGA, usually at least how many observations are required for each group? The total sample size for my research is 300. Is it enough?

[@Gaskination]

In SmartPLS, it depends on the factor with the most arrows pointing into it. The formula is 20x, where x is the number of arrows pointing into the busiest factor. So, if you have a factor that has 5 arrows pointing into it, then you need 20*5. If the busiest one has 10 arrows pointing into it, then it is 20*10. This is per group, and the arrows include indicator arrows if the factor is formative.

[@jiangpan7333]

Thanks you so much for your replay Dr. Gaskin For the number of arrows. Is it directly or indirectly pointing into the factor? Because I have some second-order constructs. If counting on the busiest second-order construct, there are 20ish items indirectly point into it..... Regards

[@Gaskination]

It is just directly. Not indirectly.

[@md.mahiuddinsabbir6125]

@@Gaskination hi there. It would be really helpful for me if you could cite a reference regarding this (sample size) even your own writings.

[@Gaskination]

@@md.mahiuddinsabbir6125 Reference mentioned here: www.researchgate.net/post/What-is-Sample-Size-Recommendations-when-using-PLS-SEM

[@moayad2055]

Please tell us how can we support your work?

[@Gaskination]

That is kind of you to ask, but I just do this as part of my service to the community. I'm happy to do it.

[@gifarireihan4821]

TYVM sirr, that's what i need to solve my prob.

[@gianluca1611]

Hi Prof, thank you for your videos. They are helping me a lot with my master thesis. I have a question: in my model, kinda similar to UTAUT, I have 4 latent variables predicting attitude and intention. I also have 3 moderating variables (Age, gender and Experience) that moderate the relationship of 3 of the lV with the attitude variable. If I am not wrong, I can test the moderating effect of gender by doing a MGA with male and female. - But how can I test the moderating effect of Age for example? I have 4 levels of Age. Should I compare two groups per time with all the possible combinations? - How can I test the moderating effect of experience (which is not a categorical variable but a continuous one varying between -7 and +7) ? Thank you very much for your help!

[@Gaskination]

Use interaction moderation instead: kzbin.info/www/bejne/hp-zl4J9qdFgosU

[@msryryka]

Hi Professor, I noticed you used "1000" for bootstrapping, is this allowed eventhough your sample is around 270ish? at first I thought we need to put the DF in that section. Thank you :)

[@Gaskination]

This is the number of times it will run the analysis with replacement. The recommended is 5000 (in most recent literature). I just left it at 1000 as a default. I usually don't find any additional benefit to use any more than 2000.

[@kshiry003]

Thankyou Dr for your multiugroup analysis tutorial. I want to use age as a moderating variable and i have 4 groups. Is it possible for me to do multi group analysis or i should follow typical way of moderation.

[@Gaskination]

Four groups cannot be compared simultaneously. You would have to compare them in pairs. Or you can say that two groups are young and two groups are old.

[@kshiry003]

James Gaskin thankyou dr

[@afaadd4905]

Dear sir hope well. i analyzed my data in PLS 3.2.7 by following your vids but i faced issued in MGA because it does not analysis indirect specific effect, others words , it does not analysis mediation effect in MGA. My study model contains two mediation hypothesis, which are not measuring by MGA, so what i do. Even you can check it, there is not any specific indirect effect option in MGA.

[@Gaskination]

That's unfortunate. The next best thing you could do is to split the data into two datasets (based on the groups for MGA). Then try the same model with each dataset. If you don't run the MGA, then it should produce the specific indirect effects for the model. Then you can compare these two different groups, but you won't have a statistic for their difference.

[@TankAqil]

Hi, James! Thank you for your videos. They are really helpful :) I have a quick question: So, can we use these results if the Frequency was a moderator?

[@Gaskination]

Agil Valiyev If frequency is a grouped variable (e.g., 0=low frequency; 1=high frequency) then MGA is a good match. If instead, frequency is on a scale (e.g., 0=never, 1=occasionally, 2=sometimes, 3=often, 4=always), then an interaction is probably a better approach. You can find my video on PLS interactions here: kzbin.info/www/bejne/hp-zl4J9qdFgosU

[@TankAqil]

@@Gaskination Thank you very much for replying :) So, if I understood correctly, for gender, MGA is okay to implement? :)

[@Gaskination]

@@TankAqil Yes, since Gender implies groups, multigroup analysis is a good match.

[@TankAqil]

@@Gaskination I have a quick question :) If the p-value in "Parametric Test" table is significant, do we interpret that there is a moderating effect? Or if the p-value is not significant, we can interpret it as there is no moderating effect of gender on particular relationship?

[@Gaskination]

@@TankAqil The significant p-value indicates that there is a difference between groups (i.e., there is moderation). If not significant, then there is no difference.

[@amineoulmane5659]

Super video. I recommend

[@c.w.chaturanisilva4444]

Thank you very much!

[@adisyouf]

how to analysis gender as a moderator

[@Gaskination]

As long as Gender is coded as binary in your dataset, then you can just use the steps in this video.

[@sukhyunlee3455]

Hello thank you so much for the videos~ It helps me a lot and I have some question regard to the multi group analysis:) I want to compare the 3 group and I created the 3 groups but when I was trying to do the MGA the problem is on the setting they only have the GroupA and GroupB Each group have the 3 categories that I made up then in order to compare the 3 categories do I have to select all for the GroupA and GroupB. Then isn't the result come out from the comparison of 9(3*3) Thank you so much for reading

[@Gaskination]

Unfortunately, the human mind can only compare two things at once. So, to compare three groups or more, you would have to compare them in pairs. This could be group A versus all others or it could be Group A versus B and then B versus C, etc.

[@sukhyunlee3455]

I see thank you for the kind reply :) Hope you have a great day!

[@mssaraeva007]

In case of more than 2 groups you need to perform OTG (omnibus test of group differences), which includes MGA and permutation. Unfortunately, for now it's only possible to analyse using R software. Hope it helps. As for pairwise comparison, you need to do a correction of alpha (e.g. Bonferroni correction)

[@ashpain07]

tq prof!

[@OsamaHaneya]

0:44 That was a pain.