Moderated Regression Analysis

[Music] what is a moderation effect what is a moderated regression analysis when do we use it what are underlying assumption and how to do it hi my name is fabin fosu in this video i will address these questions in this video i will talk about two things first i will provide an introduction what is moderate regression analysis and second i will also give you some ideas how to conduct a moderated regression analysis let me begin by giving you an example think of the case you have pain or you have headache what can you do you can take medicine you can take aspirin it's a medicine invented by a german company bayer in 1899 you take the medicine and your pain hopefully goes away however there's also side effects but that's not so much of interest for us now what we're interested in is moderating effects a moderating effect is something that intervenes uh and in the case of aspirin that would be alcohol if you drink alcohol and you take aspirin in combination that could result in severe stomach bleeding and other effects because you do both things in combination and that results in effect that is very different from the original effect so it's influence or moderated there are certain assumptions that are important so that we can trust the results of moderation analysis the assumptions are very similar to regressions because here in this video also we talk about moderated regression analysis more specifically yeah i'm thinking of multi-culinarity that means the independent and moderating variables are not strongly correlated and second homoscedasticity so that means also the variance of the variables are relatively equal i talk more about this in my video on regressions if you would like to know more and then for a moderate regression analysis we would also expect that there's a direct relationship between the independent and the dependent variable in the previous case taking aspirin would reduce pain the different types of moderation variables we can think of continuous variables such as h it could be from 0 to 120. in social science research we would typically convert these continuous variables into mean centered variables that would mean we would compute the mean and then subtract the mean from all variables and that would result in the mean centered variable why do we do that because we have often small medium-sized samples if you don't do that it could result in multi-culinarity so which would inflate and manipulate your results and we don't want that yeah that's why we mean center variables in social science research then we also have dichotomous or binary variables typically 0 1 just to give you an example and i talk about cats versus dogs which animals are more liked by people and that would be 001 variable then we also have categorical variables it could be the example of colors there are many different colors let's say 10 different colors which color is most liked by consumers we can't really enter such kind of variable in moderated regression analysis because that's not possible what we could do however then is we would compute dummy variables it could be you would like to investigate whether blue is a very preferred color then you would create a dummy variable 1 equals blue 0 equals all other colors and then you can enter into a moderated regression analysis i talked about the assumptions underlying regression analysis one such assumption is that the independent variable should be significantly related with the dependent variable i gave the example aspirin taking aspirin would reduce your pain and so there's a direct relationship between these variables then the moderating effect if you see the example of alcohol would influence this relationship if you drink more alcohol then that would not reduce but in combination with the consumption of the drug would increase pain in ideally this moderating effect this variable should not have any relationship with the dependent variable should there should be no significant relationship that's what we would call a clean moderating effect some statistic books even expect that this is a must criterion to be considered a moderating effect then however we also have moderating effects where then the moderating effect does not only influence the relationship between independent and dependent variable but it also has a direct relationship with the outcome variable so that's not ideal some people call it a dirty moderating effect so we see that yeah but that's usually what we do not would like to see so far i've talked about the background and assumptions of moderated regression analysis now let me start by talking about how to conduct a moderate regression analysis first i have to say there's so many different statistical programs that you can use to do so it could be spss r starter and so forth here i would like to give an overview of how to do it and what uh the basic understanding and then you can apply it uh using the program you like first we can think of the assumptions underlying regression analysis or moderating regression that is that needs to be met so i would recommend you to check for the effects of the direct effect on the outcome variable it should be statistically significant then you would probably compare that model yeah so your baseline model where you have your control your independent variables and that with a model where you add the interaction term or the moderating effect and if you compare these two models yeah it would be very important that you see that your new model including the interaction term would add variance or would add explanatory power there's a significant r square change there you would like to look at that and then you would also look at the significance effects of your interaction term of your moderating effect it should be statistically significant also what you would do is to check your your model for multi-culinarity because if there's a very high multi-clarity maybe there's some other underlying reason why we see this statistically significant relationship then also what you what i would recommend you to do is check for all the effects of the independent and moderating variables if we have a clean or a dirty moderating effect i've already talked about mean centering variables it's very important in social science research so let me pick it up again so it would be very important to mean center your independent variable and your moderating variable before you run your analysis because mean centering yeah would usually reduce collinearity between the independent variable and the moderating variable which would mean that we can trust the results more there are also other scholars that say oh mean centering oh that doesn't help we don't need it and that's also somewhat depends on your discipline and your sample in social science we very often have small or medium sized samples there if you mean center variables you you can see substantial differences and i've tried this myself several times yeah and it makes quite a difference if you have large surveys large data thousands or millions of respondents then mean centering doesn't change much in terms of disciplines finance economics they're not really big fans of mean centering whereas in social science it's very common are so many different statistical packages that you can use to do a regression and moderated regression analysis to test for moderating effects spss sars r starter are probably among the most common programs i have another video where i talk about the pros and cons of these different statistical packages i don't really endorse one but if your main purpose is to test for moderating effects and you do it more in a let's say exploratory fashion there's one tool i think that is extremely helpful that allows you to conduct many moderating effects with just almost just one click so it's so easy and that's the process macro and you can download it for spss sas up and r and it's so handy because it does the mean centering automatically you don't need to compute variables and you can also just throw in many many variables and the program will test everything for you automatically one click and the program does almost everything for you great you have conducted your statistical analysis now let us have a look at the outputs what are the numbers you would be interested in obviously you would be interested in the p-value of significance of your moderating effect if it's below 0.05 which is the common standard then we would consider the statistical significant moderating effect we also would like to look into the b the coefficient therefore the effect is it's small large and also to understand the effects a little better we would like to look into the r square change i remember the model without the moderating effect and compare that model after entering the moderating effect what's the change in r square how much more variance can the second model explain more than the first model nowadays also the typical expectation for many journals would be that you would conduct a simple slope comparison so you would compare you know the values of the moderating effect at low level with a high level typically plus minus one standard deviation also if you use these statistical problems they can compute it automatically for you or you can also compute that by hand by entering these variables with plus and minus one standard deviation and what is also very helpful to illustrate your results through a figure that will help you to better understand your results let me give you an example of the graphical illustration we call it the interaction plot yeah have a look at my interaction plot from one of my recent papers i'm a professor of business that's a business context so the dependent variable is knowledge transfer by repatriates repatriates are people who have worked in a foreign country and then return to the headquarters of that company and then they share the knowledge that they acquired from working abroad in the headquarters yeah so it would be important for them to be embedded embedded and highly well integrated in the structures and headquarters and then the moderating effect would be communication frequency with the former colleagues in the foreign country the more they communicate and the more embed it the more knowledge they can transfer you can see that also the slope well the dotted slope that's upward slope right it's a significant relationship and then minus one standard deviation would be if they have only little or low communication frequency and then you can see the line is almost flat okay so then there's no statistical significant relationship so we would say that the level of knowledge transfer and embeddedness fit in the headquarters depends on the communication frequency low versus high communication frequency makes a difference and that would be significant moderating effect in this video i talked about moderated regression analysis moderate regression analysis can enhance understanding of relationships between variables and typical questions would be when under what boundary conditions can some effect mitigate increase reduce strengthen the relationship or the effect on certain outcome variables so it's a very powerful and very helpful analysis to better understand relationships thus it is no surprise that moderate regression analysis is a very common statistic analysis and social science research i would also recommend you to consider in your research design and in your data analysis i think we've reached the end of this video i hope it was helpful to you i hope it didn't cause any uh headache or any pain and maybe you even feel like enjoying some drinks cheers [Music]

Channel: Prof. Fabian Froese

Views: 449

Rating: 5 out of 5

Keywords:

Id: VAQNES85hCs

Length: 14min 50sec (890 seconds)

Published: Fri Feb 12 2021