Moderator analysis

all over the aim of this demonstration is to show you how to do a moderation analysis in SPSS in this case the dependent variable is word count and I have two independent variables GSR and experience they represent the physiological arousal in response to a set of musical stimuli in experience is an encoding of level of previous exposure to the type of music in these stimuli so a word count is for response measure it's a it's verbal report of how how many emotions people feel in response to the items of musics I'm hypothesizing that physiological arousal will enable me to predict responsiveness verbal responsiveness and so will previously experience of that type of music I'm supposing that experience may or may not have an effect on how easy it is for people to verbalize their emotions and response to the music so the simplest sort of analysis would be to do a linear regression and put the word count variable as dependent on these two is independence and if I do that I will see that there is a model the model is significant I mean but neither of the two independent variables is independently significant and in case that might be an instance of multicollinearity I can do Swift correlation and see whether that is the case but it's not there's no that's not the explanation it may well be therefore that there is some other effects and I'm going to test to this by doing a moderation analysis in order to do this I need to first centralize the two variables and then multiply them together in order to get a third variable which I will use to test for moderation there is a simple way to to do the centralization bit which is merely to centralization describes the process of getting the finding the mean of the variable and then subtracting that mean from every value of the variable so you make a new variable whose mean is zero well in fact multiplication by a constant value is not going to affect the analysis of this so what I can do in fact rather than centralized I can actually take it I can standardize the variable and that will do the same thing and so I'm going to actually put these variables in here if I go into the descriptives menu I have a function which will enable enable me to save those values as variables and I can see that that's now that's not happened and you know the last thing I've got to do is to compute a variable which is the product of those two so I'm just going to call it moderator I'm going to make it equal to the product of the of the two sets chords and that has produced a moderator variable if I put all those now into the into the analysis so I add moderator variable I will now see something interesting first of all the model summary is now predicting 37 that's nearly 38% of the variance just for comparison the previous one predicted 13% so it looks as though we're doing the right thing now looking at the three variables GSR has dropped out interestingly the variable is very important but only via its moderator component in fact it's so non significant now I'm tempted just to get rid of it and see what happens with a subsequent analysis which just says - and you experienced the moderator variable significant we've actually increased the ingested r-square to 39 and a half percent so that shows that moderation is going on now how to interpret it if I do some graphs of the responsiveness against the GSR stimulus although GSR is not appearing as an independent predictor here to any significant extent it is still influencing the value of the word counter there the verbal responsiveness and I'll show you how that is happening now if we remember that what moderation is moderation is when you have the effect one variable on the DV depends on the level of the other variable it's rather like the interaction effect in a ver and to show this most dramatically I'm going to actually look at the three values this experience fair variable actually was coded fairly simple mindedly only three valid values 1 2 and 3 I'm gonna show you what happens when we look at those three variables those three levels separately and then we look at what GS are but the relationship between GSR and and the DB so I'm going to split the file into separate values depending on the level of experience which I do this now when I go back and actually going to do a scatter plot here to illustrate the relationship between word count and GSR this shows you the relationship GS are the physiological responses down the bottom and then this word count is if you like cognitive verbal response and you can see that through the three values of experience it's very different for the people who hadn't had much of any exposure to classical music it seems that responsiveness that the stronger the physiological arousal was the less verbal they were in their description the music I know it's a it's a little bit of an outlier down here but if we look at experience level to moderate experience there's really nothing going on there there's another little individual here but if we look at now level three we see that the relationship between responsiveness and word can't is going completely the opposite direction to what it was in experience one and that is really what is giving us the large effect of of the moderator variable it's showing us that GSR on its own doesn't tell us anything because the effect of GSR on the TV is totally opposite depending on whether you're dealing with people who have prior experience to music or people with very little or no experience to music so if you just do a simple multiple regression you will lose the you'll lose out on the effect of GSR it it is important through its effect its interaction effect on musical experience and if you do a moderator analysis like this you will be able to pull apart those two those two effects thank you

Channel: Rory Allen

Views: 342,148

Rating: 4.7115841 out of 5

Keywords: statistics

Id: sj2XWfZaxek

Length: 8min 38sec (518 seconds)

Published: Tue Feb 14 2012