I am new to Smart PLS so I am unfamiliar with the terms, I would like to know what P values mean and why they are not very dependable information? And also your videos are really fascinating! Keep up the good work!
p values are the probability that nothing is going on - that there is no effect. So, a p-value of 0.05 indicates a 5% chance that there is no effect (or a 95% confidence that there is an effect). For interactions, the p-values tend to be a lesser indicator than the slopes of the separate effects. However, we cannot ignore the p-value, as it is an indication of a reduction in error. So, even if we see a diverging set of slopes, we might have too much error in those slopes to be confident in their direction. In such a case, a multigroup analysis could be done with low/high groups on the moderator to see if the individual slopes were statistically significant. If they are, then we can compare them.
Hello James, Thanks a lot for all your awesome and helpful videos! May I ask about the Job category, Gender and Experience variables? If I want to consider them as control variables, what should I do to test their effects on satisfaction with customers?
Because job category and gender are both coded as binary in this dataset, and experience is just continuous, you could simply include them as regular predictors.
Professor James, great vídeos, and tutorials as well. I've accompanying your work, and I'm grateful for the amazing contributions you've done. Concerning moderation effect, I have a question about the connection of the moderating variable to the dependent variable. In Smart PLS 4, when calculating the moderating effect, do we have to connect the moderating variable to the dependent variable too, as we connect to the "arrows" to be moderated? I saw another tutorial video where in Smart PLS4 it was only necessary to connect the moderating variable to the arrows that represent the relationship being moderated. Another question is that if we are running a reflective model, being moderated by some variable, do we have to use PLS algorithm, or consistent PLS algorithm, Bootstrapping or consistent Bootstrapping? Best regards!
It is conventional to have a direct effect from the moderator to the DV. This is how interactions are typically set up. As for PLSc, it is used when all factors are reflective. However, if it gives you erroneous results, you can switch back to PLS. This still seems to be an issue sometimes.
Hello Dr, excuse me, do you have slope analysis for standardized regression weight for moderating effect ?? I have run moderating interaction in amos, and I need slope analysis based on standardized estimate, of course I have searched Jeremydawson website, but I have not found it. many thanks in advance.
Dear Professor, I am dealing with a complicated model of multi mediator and moderator analysis , here the model is like, x1 -> moderators and mediators -> x2- y1. Now the results are showing than, connection specific indirect effect from x1 to y1 is insignificant but there is no way I could see any specific connection result from X1 to X2 the excel result file of PLS4 analysis. The only results it is showing is from X1 to Y2. But nothing about the direct or indirect specific analysis with a specific moderator from X1 to X2. Can you please tell me where should I pay attention to find it from the excel results. Thank You.
Thank you for your helpful video. I have a question. If the p-value of the moderating effect is too big, can I get rid of the hypothesis on the moderating effect? or as you said 'do not concentrate on p-value when it comes to moderating effect', can I analyze the moderating effect with simple slopes plot, ignoring p-values? I am curious about how people deal with this problem especially when it comes to paper writing.
I don't think I understand. Excel does not need to be involved to test and report moderation. You can just report the slopes plots and the table of estimates.
Prof James Gaskin, hope you are doing good. I have my Ph.D. research model comprising one independent variable, two dependent variables, three mediators, and two moderators. Using smart pls 4, should I run separate models for direct relationships, mediating relationships, and moderating/moderated mediated/mediated moderated relationships, or must I run this complete model at one time as some results are totally different when running separate models and when running it as a whole model?
I just want to now the moderation method that used in smartpls 4. as we know that there are product indicator, two stage approach and ortogonalization. which one of those method used in smartpls 4?
Are there scientific articles or books you can recommend that talk about how we should not rely on p-value as much and more on simple slope analysis for moderation effect?
P-values are commonly used as a way to test statistical significance. They reflect the likelihood of obtaining a result as extreme as, or more extreme than, the observed result if the null hypothesis were true. In the context of moderation analysis, the null hypothesis is usually that there is no moderation effect. If the p-value associated with the interaction effect is small (e.g., p < .05), you might reject the null hypothesis and conclude that the moderation effect is statistically significant. However, p-values have their limitations. They are influenced by the sample size, and they do not convey information about the magnitude or the practical significance of the effect. Moreover, relying solely on p-values can lead to dichotomous thinking (i.e., significant vs. non-significant), which has been criticized in the statistical literature (see Wasserstein, R. L., & Lazar, N. A. (2016). "The ASA's statement on p-values: context, process, and purpose." The American Statistician, 70(2), 129-133). Graphical methods such as simple slopes plots and floodlight analysis can complement p-values and provide additional insights. Simple slopes plots can help visualize the interaction effect and the nature of the moderation. They show the relationship between the predictor and the outcome at different levels of the moderator. This can be particularly helpful when the interaction effect is complex and not easily interpretable from the coefficients alone. Floodlight analysis (also known as Johnson-Neyman technique) goes a step further by identifying regions of significance: values of the moderator where the effect of the predictor on the outcome is significant. This can provide more nuanced insights than a single p-value for the interaction term. Spiller, S. A., Fitzsimons, G. J., Lynch Jr, J. G., & McClelland, G. H. (2013). "Spotlights, floodlights, and the magic number zero: Simple effects tests in moderated regression". Journal of Marketing Research, 50(2), 277-288. In sum, both p-values and graphical methods have their place in moderation analysis. P-values provide evidence about the existence of a moderation effect, but they are not sufficient to understand the nature or the magnitude of the effect. Graphical methods can provide these additional insights and should be used in conjunction with p-values. This balanced approach aligns with the recommendations of statisticians and methodologists (e.g., Hayes, A. F. (2013). "Introduction to mediation, moderation, and conditional process analysis: A regression-based approach". The Guilinary Press).
Professor James, So my model tests moderation effect of gender on one specific path. first thing i wish to have clarity on is, is it okay to simply show results of interaction effect or should i move to MGA ? Further the results show insignificant moderation. The slopes intersect each other in the very initial stage but then move positively upward. i am confused regarding how to interpret it. if there is any literature regarding how interction effect is suitable choice kindly refer. i shall be very grateful to you.. thank you
Using an interaction, instead of MGA, is okay if it is binary. If it is multinomial (more than two categories), then MGA is required. Usually if there is an intersection of slopes, then this means there is moderation (unless they only very slightly diverge). I would rely more on the slopes than on the p-value.
@@Gaskination thank you so much for the reply Sir. so the slopes intersect each other initially and then positively diverge. P value shows insignificant relationship. how can i interpret these results ? i would be really grateful for your guidance. Thank You.
@@sochtosach6861 The limitation here is that the p-value is non-significant, implying that those slopes might not represent true (consistent) slopes. You might benefit from testing this in a multigroup analysis to see if each individual slope is significant. If it is, then you could rely on this interaction. If not, then there is probably too much error to distinguish consistent slopes.
Hey James. I've got a question to my moderation analysis, hope you can help. My moderation effect (moderator=intolerance for crowding) is insignificiant and the beta is positive, but quite low (b=0.014, p=0.416). The relationship between my indepedent variable (perceived human crowding) and dependent variable (satisfaction with the shoppting trip) is negative (b=-0.389) and significant. Also, intolerance for crowding has a negativ insignificant effect on satisfaction (b=-0-071, p=0.237). So how do I interpret this? Intolerance for crowding is not significantly moderating the relationship. But does my moderator dampen the negativ relationship between crowding and satisfaction? It doesent really makes sense to me. Kind regards
Hi James, thank you so much for your immensely helpful videos! I was wondering if you could provide guidance on the interpretation of a slope graph that I could not find in your video. We are analyzing a negative main effect with a moderator that we hypothesized to weaken the main effect. PLS-SEM revealed a negative β for the moderator, which we initially interpreted as confirming the weakening effect of the moderator on the negative main effect. However, the slopes plot shows that the line for higher levels of the moderator (+1 SD in the graph) is much steeper than the others. Doesn’t this mean that the moderator strengthens the main effect? We are now wondering how we can best interpret this seemingly contradicting result, or if we are making a logical error?
Negatives are always confusing. When you have a negative direct effect that is being moderated, a "weakening effect" would bring the effect closer to zero (rather than more negative). A strengthening effect would bring it away from zero (i.e., make it more negative). So, if the negative slope was steeper for +1sd, then it is strengthening the effect (however, this might be what you intended when you theorized it). Hope this clarifies.
Hello James, I have 3 questions, please 1) Does SmartPLS 4 calculate the Q square effect Size? or it should be done manually like SmartPLS 3? 2) For the Q square calculation (predictive relevance), should I remove my moderation effect & control variables? OR run the whole model for Qsqaure altogether 3) one of the direct relationships is not significant (supervisor support doesn't have a relationship with work-life balance). However, when I added the moderator (Self-Efficacy) the whole relationship including the interaction became significant (SS*SE*WLB). in this case shall I accept or reject the moderating Hypothesis?
Hi James! Firts of all, I would like to thank you for your videos! I have a question reguarding the order of analysis in PLS-SEM when there is a moderator. Firts, we assess the measurement model, then the structural model and only after that we asses the moderator, right? But my big question is: when do I include the moderator in the model? Does it need to be present on the assessment of the measurement model? Or only in the structural? Or do I make and additional analysis and I evalute first the measurement and structural model without the moderator included? I hope you can help me!
Yes, assess measurement then structural model. As for when to include the moderator, if it is latent, then it should be evaluated the same as any latent factor in the measurement model. If it is a grouping variable, then it can be brought in afterward just to test moderation.
The results of simple slope are different when there is only one moderating path and when there are more than one moderating path. How to understand and how to proceed?
Hello James Gaskin, thanks a lot for all your awesome and helpful videos!!! I have been able to clarify many questions and am totally thankful for that. Nevertheless, I am having difficulties with my data analysis and/or model building. I would be very happy if you could help me by answering my question. I have been working on the acceptance of voice assistants as part of my Bachelor thesis and have conducted an empirical survey based on the UTAUT 2 model. In the UTAUT2 paper the authors say, Use, was modeled using six formative indicators. Does this mean that I need to treat use as a formative construct in my model? My second question is about the moderators. The authors say that significant path coefficients were found with all higher-order interaction terms - how can I measure the influence of the moderators gender, age, and experience all at once? Like Age*Gender*Experience*HABIT? I would be so incredibly grateful if you could help me! Love and thanks for the great content!
Yes, if they modeled it formatively, then you should model it formatively. As for the three moderators, I would recommend you keep them separate, or moderate by gender as a multigroup variable while you have one interaction term. Three and four way interactions become very difficult to interpret and decipher what is creating the effect. One way around this, if you have a very large sample size, is to create a grouping variable that represents high and low values of each possible combination; e.g., young male with few years of work experience, young female with many years of work experience, etc. You may also find that age and work experience are highly overlapping, so that one is redundant with the other and not necessary to add value to the model.
@@Gaskination Can I just hang in here^^ What would be your rule of thumb when you have 5-10 potential moderators to check? I would check for similarities in the groups first (e.g. age and work experience), to reduce number of potential moderators. However, if I have still 5 parameters (e.g. age, gender, nationality, wealth and housing type) left, could I create a group variable with 3 parameters (e.g. young, male, nationality A VS. old, female, nationality B)? Or what would you suggest? Thank you again!
@@ragwsjuve856 I would argue that most of these are control variables, not moderators. The moderator is only something you are theoretically interested in comparing. If it is just a potential confound, include it as a control, not a moderator.
This is probably the most interesting because it shows that the IV has no effect on the DV except when interacting with the moderator. Thus, the moderator is needed to observe an effect.
SmartPLS just reports the absolute value of the division. Amplitude (or distance from zero) is what matters with the t-stat (although something can be inferred from direction - but we can already get this from the Beta O or M).
Dear Dr James, is it necessary to have a direct effect significant! to consider moderating effect also significant too! Example: the direct effect of moderating variable to dependant variable is not significant but the moderating effect is significant....whats your say on this situation? Kindly answer
It is not required. It is often the stronger argument for moderation when the direct effect is not significant. That means the IV has no effect on the DV unless we consider the moderator.
The improvement in moderation modeling in smartpls4 is soooooo coooooool
Hello doctor. Best videos i have ever seen online. Can u share the demo data to do this analysis?
This data is available on the homepage of statwiki.gaskination.com/ It's the Sohana dataset
I am new to Smart PLS so I am unfamiliar with the terms, I would like to know what P values mean and why they are not very dependable information? And also your videos are really fascinating! Keep up the good work!
p values are the probability that nothing is going on - that there is no effect. So, a p-value of 0.05 indicates a 5% chance that there is no effect (or a 95% confidence that there is an effect). For interactions, the p-values tend to be a lesser indicator than the slopes of the separate effects. However, we cannot ignore the p-value, as it is an indication of a reduction in error. So, even if we see a diverging set of slopes, we might have too much error in those slopes to be confident in their direction. In such a case, a multigroup analysis could be done with low/high groups on the moderator to see if the individual slopes were statistically significant. If they are, then we can compare them.
Thank you very much, my dear🙏🙏🙏🙏🙏
Woauh!! I am so happy to see that! Thank you
Great sir 👍❤
Thanks it is an important video I need. tx
great Video and feature!
Thanks James !
Hello James,
Thanks a lot for all your awesome and helpful videos!
May I ask about the Job category, Gender and Experience variables? If I want to consider them as control variables, what should I do to test their effects on satisfaction with customers?
Because job category and gender are both coded as binary in this dataset, and experience is just continuous, you could simply include them as regular predictors.
Professor James, great vídeos, and tutorials as well. I've accompanying your work, and I'm grateful for the amazing contributions you've done. Concerning moderation effect, I have a question about the connection of the moderating variable to the dependent variable. In Smart PLS 4, when calculating the moderating effect, do we have to connect the moderating variable to the dependent variable too, as we connect to the "arrows" to be moderated? I saw another tutorial video where in Smart PLS4 it was only necessary to connect the moderating variable to the arrows that represent the relationship being moderated. Another question is that if we are running a reflective model, being moderated by some variable, do we have to use PLS algorithm, or consistent PLS algorithm, Bootstrapping or consistent Bootstrapping? Best regards!
It is conventional to have a direct effect from the moderator to the DV. This is how interactions are typically set up. As for PLSc, it is used when all factors are reflective. However, if it gives you erroneous results, you can switch back to PLS. This still seems to be an issue sometimes.
Hello Dr, excuse me, do you have slope analysis for standardized regression weight for moderating effect ??
I have run moderating interaction in amos, and I need slope analysis based on standardized estimate, of course I have searched Jeremydawson website, but I have not found it.
many thanks in advance.
Dear Professor, I am dealing with a complicated model of multi mediator and moderator analysis , here the model is like,
x1 -> moderators and mediators -> x2- y1. Now the results are showing than, connection specific indirect effect from x1 to y1 is insignificant but there is no way I could see any specific connection result from X1 to X2 the excel result file of PLS4 analysis. The only results it is showing is from X1 to Y2. But nothing about the direct or indirect specific analysis with a specific moderator from X1 to X2. Can you please tell me where should I pay attention to find it from the excel results. Thank You.
Perhaps this is the video you're looking for: ruclips.net/video/W3awqjXRhBo/видео.htmlsi=-B2ipmg_pMifuSYp
Thank you for your helpful video. I have a question. If the p-value of the moderating effect is too big, can I get rid of the hypothesis on the moderating effect? or as you said 'do not concentrate on p-value when it comes to moderating effect', can I analyze the moderating effect with simple slopes plot, ignoring p-values? I am curious about how people deal with this problem especially when it comes to paper writing.
Here is a useful discussion on this point: chat.openai.com/share/ba3336ee-a6f3-4d05-b606-ed8b89bb6607
thank you so much!
Thank you for the video. Which excel sheet should i use to report the moderation slopes? as i have three iv with one moderator (ordinal).
I don't think I understand. Excel does not need to be involved to test and report moderation. You can just report the slopes plots and the table of estimates.
Prof James Gaskin, hope you are doing good. I have my Ph.D. research model comprising one independent variable, two dependent variables, three mediators, and two moderators. Using smart pls 4, should I run separate models for direct relationships, mediating relationships, and moderating/moderated mediated/mediated moderated relationships, or must I run this complete model at one time as some results are totally different when running separate models and when running it as a whole model?
It is best to run altogether, although you might run the moderators separately if they are grouping moderators.
I just want to now the moderation method that used in smartpls 4. as we know that there are product indicator, two stage approach and ortogonalization. which one of those method used in smartpls 4?
I'm pretty sure it uses the two stage approach.
Are there scientific articles or books you can recommend that talk about how we should not rely on p-value as much and more on simple slope analysis for moderation effect?
P-values are commonly used as a way to test statistical significance. They reflect the likelihood of obtaining a result as extreme as, or more extreme than, the observed result if the null hypothesis were true. In the context of moderation analysis, the null hypothesis is usually that there is no moderation effect. If the p-value associated with the interaction effect is small (e.g., p < .05), you might reject the null hypothesis and conclude that the moderation effect is statistically significant.
However, p-values have their limitations. They are influenced by the sample size, and they do not convey information about the magnitude or the practical significance of the effect. Moreover, relying solely on p-values can lead to dichotomous thinking (i.e., significant vs. non-significant), which has been criticized in the statistical literature (see Wasserstein, R. L., & Lazar, N. A. (2016). "The ASA's statement on p-values: context, process, and purpose." The American Statistician, 70(2), 129-133).
Graphical methods such as simple slopes plots and floodlight analysis can complement p-values and provide additional insights. Simple slopes plots can help visualize the interaction effect and the nature of the moderation. They show the relationship between the predictor and the outcome at different levels of the moderator. This can be particularly helpful when the interaction effect is complex and not easily interpretable from the coefficients alone.
Floodlight analysis (also known as Johnson-Neyman technique) goes a step further by identifying regions of significance: values of the moderator where the effect of the predictor on the outcome is significant. This can provide more nuanced insights than a single p-value for the interaction term. Spiller, S. A., Fitzsimons, G. J., Lynch Jr, J. G., & McClelland, G. H. (2013). "Spotlights, floodlights, and the magic number zero: Simple effects tests in moderated regression". Journal of Marketing Research, 50(2), 277-288.
In sum, both p-values and graphical methods have their place in moderation analysis. P-values provide evidence about the existence of a moderation effect, but they are not sufficient to understand the nature or the magnitude of the effect. Graphical methods can provide these additional insights and should be used in conjunction with p-values. This balanced approach aligns with the recommendations of statisticians and methodologists (e.g., Hayes, A. F. (2013). "Introduction to mediation, moderation, and conditional process analysis: A regression-based approach". The Guilinary Press).
@@Gaskination Thank you so much. I really enjoy your videos and wish you were teaching at my school
Professor James,
So my model tests moderation effect of gender on one specific path. first thing i wish to have clarity on is, is it okay to simply show results of interaction effect or should i move to MGA ? Further the results show insignificant moderation. The slopes intersect each other in the very initial stage but then move positively upward. i am confused regarding how to interpret it. if there is any literature regarding how interction effect is suitable choice kindly refer. i shall be very grateful to you..
thank you
Using an interaction, instead of MGA, is okay if it is binary. If it is multinomial (more than two categories), then MGA is required. Usually if there is an intersection of slopes, then this means there is moderation (unless they only very slightly diverge). I would rely more on the slopes than on the p-value.
@@Gaskination thank you so much for the reply Sir.
so the slopes intersect each other initially and then positively diverge. P value shows insignificant relationship. how can i interpret these results ? i would be really grateful for your guidance. Thank You.
@@sochtosach6861 The limitation here is that the p-value is non-significant, implying that those slopes might not represent true (consistent) slopes. You might benefit from testing this in a multigroup analysis to see if each individual slope is significant. If it is, then you could rely on this interaction. If not, then there is probably too much error to distinguish consistent slopes.
Hey James. I've got a question to my moderation analysis, hope you can help.
My moderation effect (moderator=intolerance for crowding) is insignificiant and the beta is positive, but quite low (b=0.014, p=0.416).
The relationship between my indepedent variable (perceived human crowding) and dependent variable (satisfaction with the shoppting trip) is negative (b=-0.389) and significant.
Also, intolerance for crowding has a negativ insignificant effect on satisfaction (b=-0-071, p=0.237).
So how do I interpret this? Intolerance for crowding is not significantly moderating the relationship. But does my moderator dampen the negativ relationship between crowding and satisfaction? It doesent really makes sense to me.
Kind regards
Check the simple slopes plot for the interpretation. Interactions rely less on p-values and more on changes in slopes.
@@Gaskination Thanks! I dont really know how to interpret it. Because the lines are so close and similar..
@@susanna5254 That means there is no moderation.
Hi James, thank you so much for your immensely helpful videos! I was wondering if you could provide guidance on the interpretation of a slope graph that I could not find in your video. We are analyzing a negative main effect with a moderator that we hypothesized to weaken the main effect. PLS-SEM revealed a negative β for the moderator, which we initially interpreted as confirming the weakening effect of the moderator on the negative main effect. However, the slopes plot shows that the line for higher levels of the moderator (+1 SD in the graph) is much steeper than the others. Doesn’t this mean that the moderator strengthens the main effect? We are now wondering how we can best interpret this seemingly contradicting result, or if we are making a logical error?
Negatives are always confusing. When you have a negative direct effect that is being moderated, a "weakening effect" would bring the effect closer to zero (rather than more negative). A strengthening effect would bring it away from zero (i.e., make it more negative). So, if the negative slope was steeper for +1sd, then it is strengthening the effect (however, this might be what you intended when you theorized it). Hope this clarifies.
Thank you so much for your response! That helps to clarify.
Thank you James. When you get graphs which cross each other what is the implication?
That implies strong moderation if the slopes are very different.
Hello James, I have 3 questions, please
1) Does SmartPLS 4 calculate the Q square effect Size? or it should be done manually like SmartPLS 3?
2) For the Q square calculation (predictive relevance), should I remove my moderation effect & control variables? OR run the whole model for Qsqaure altogether
3) one of the direct relationships is not significant (supervisor support doesn't have a relationship with work-life balance). However, when I added the moderator (Self-Efficacy) the whole relationship including the interaction became significant (SS*SE*WLB). in this case shall I accept or reject the moderating Hypothesis?
1. ruclips.net/video/OONQ6nhwt0s/видео.html
2. see above video
3. Moderation is supported in this case.
@@Gaskination thanks alot. Can yor recommend to me any empirical support that I can cite for the accepting the moderating hypothesis
@@ingyselim2760 statwiki.gaskination.com/index.php?title=References#Moderation_and_Multigroup
@@Gaskination thanks a million
Hi James! Firts of all, I would like to thank you for your videos! I have a question reguarding the order of analysis in PLS-SEM when there is a moderator. Firts, we assess the measurement model, then the structural model and only after that we asses the moderator, right? But my big question is: when do I include the moderator in the model? Does it need to be present on the assessment of the measurement model? Or only in the structural? Or do I make and additional analysis and I evalute first the measurement and structural model without the moderator included? I hope you can help me!
Yes, assess measurement then structural model. As for when to include the moderator, if it is latent, then it should be evaluated the same as any latent factor in the measurement model. If it is a grouping variable, then it can be brought in afterward just to test moderation.
@@Gaskination Ok! Thank you so much for your help!!
The results of simple slope are different when there is only one moderating path and when there are more than one moderating path. How to understand and how to proceed?
Yes, every time you add or take away a path in SEM, everything can change. It is an interconnected network of interdependent effects.
@@GaskinationHow to perform moderation analysis correctly? One path or multiple paths at a time?
@@user-bh4ft3vz6r Multiple paths is more closely approximating reality, but sometimes sample size requires us to do just one at a time.
Thank you for the video Sir.
If u could please share some paper wherein moderation graph has been explained this way
Preacher, K. J., Curran, P. J., & Bauer, D. J. (2003). Simple intercepts, simple slopes, and regions of significance in MLR 2-way interactions.
@@Gaskination thank you so much Sir
Hello James Gaskin,
thanks a lot for all your awesome and helpful videos!!!
I have been able to clarify many questions and am totally thankful for that.
Nevertheless, I am having difficulties with my data analysis and/or model building.
I would be very happy if you could help me by answering my question.
I have been working on the acceptance of voice assistants as part of my Bachelor thesis and have conducted an empirical survey based on the UTAUT 2 model.
In the UTAUT2 paper the authors say, Use, was modeled using six formative indicators.
Does this mean that I need to treat use as a formative construct in my model?
My second question is about the moderators. The authors say that significant path coefficients were found with all higher-order interaction terms - how can I measure the influence of the moderators gender, age, and experience all at once?
Like Age*Gender*Experience*HABIT?
I would be so incredibly grateful if you could help me!
Love and thanks for the great content!
Yes, if they modeled it formatively, then you should model it formatively. As for the three moderators, I would recommend you keep them separate, or moderate by gender as a multigroup variable while you have one interaction term. Three and four way interactions become very difficult to interpret and decipher what is creating the effect. One way around this, if you have a very large sample size, is to create a grouping variable that represents high and low values of each possible combination; e.g., young male with few years of work experience, young female with many years of work experience, etc. You may also find that age and work experience are highly overlapping, so that one is redundant with the other and not necessary to add value to the model.
@@Gaskination thank you so much for you answer!
@@Gaskination Can I just hang in here^^
What would be your rule of thumb when you have 5-10 potential moderators to check?
I would check for similarities in the groups first (e.g. age and work experience), to reduce number of potential moderators.
However, if I have still 5 parameters (e.g. age, gender, nationality, wealth and housing type) left, could I create a group variable with 3 parameters (e.g. young, male, nationality A VS. old, female, nationality B)? Or what would you suggest?
Thank you again!
@@ragwsjuve856 I would argue that most of these are control variables, not moderators. The moderator is only something you are theoretically interested in comparing. If it is just a potential confound, include it as a control, not a moderator.
@@Gaskination Thank you! :)
Hi James, what about cross over interaction (insignificant main effect, significant interaction effect), what we can say ? thank you
This is probably the most interesting because it shows that the IV has no effect on the DV except when interacting with the moderator. Thus, the moderator is needed to observe an effect.
Why T never shows negative? When the effects are negative?
SmartPLS just reports the absolute value of the division. Amplitude (or distance from zero) is what matters with the t-stat (although something can be inferred from direction - but we can already get this from the Beta O or M).
@@Gaskination Thank you for ur answer
Dear Dr James, is it necessary to have a direct effect significant! to consider moderating effect also significant too!
Example: the direct effect of moderating variable to dependant variable is not significant but the moderating effect is significant....whats your say on this situation?
Kindly answer
It is not required. It is often the stronger argument for moderation when the direct effect is not significant. That means the IV has no effect on the DV unless we consider the moderator.
@@Gaskination thank you Dr.