Moderation and Mediation

so webcast is going to have a look at moderation and mediators thing you need to do is go to Andrew Hayes's website and download his process tool so at the moment this is stored on the website for his new book called introduction to mediation moderation and conditional process and analysis which incidentally is a fantastic book and I heartily recommend that you get a copy and it's also very very good way to follow up my book chapter on mediation because mine do some basic stuff but his book talks about much more complex and models and things like that and goes into a lot more detail this great book so if you can find this website and then you just scroll down to this section here where it says process for SPSS and SAS and you download this zip file so just click on the download it will download your computer somewhere and then unzip that file and then you'll have the file that you need to load into SPSS and that's what we'll have a look at now if you're on Windows what you need to do is open SPSS as an administrator now this is really important if you don't run it as an administrator none of this will work so in Windows you start off or you would start off finding SPSS in your in your kind of programs and it might not necessarily be called version 20 I just happen to have version 20 on my PC that I use for getting PC screenshots and so you click on the right click on that and then you'll get this dialog box here which normally you know you might just sort open it but we want to select this thing here run as an administrator so if you click on that I mean nothing much will change SPSS will kind of open up in the normal way but it opens up and giving SPSS access to your computer so allows it to kind of change things and that's important because when we load this process tool SPSS needs to change things on your system so you've got to run it as an administrator you such a long time since I use Windows a contra member it may throw up some dialog boxes saying are you sure you want to London's administrator or do you want to let SPSS wreak havoc on your computer or whatever and the answer to that is yes you do SPSS needs to wreak havoc on your computer so entering SPSS what you need to do is go to the utilities menu and then select custom dialog box sorry custom dialogs and then you want to select install custom dialog so if you click on that you'll get a standard dialog box that opening files and I start the process tool on my desktop but basically you just need to go and find this file so this is the one of the files when you're unzipped the zip file from Andrew Hayes's website one of the files within it is this thing here process SPD SPD and you just open that and well I'm getting an error message because I've already had already previously installed it on my system but basically what will happen is that will then install the dialog box for you so there you go it's installed out I've just overwritten my previous version so then what happens is in the analyze menu now when you go to regression what you'll find is there is a new a new additional item in your menu so this wouldn't have been there before and it's it's only there because we've now installed this custom dialog box and that is the process menu by Andrew Haynes so that's what we'll use to run our moderation and mediation models actually will deal with moderation first so the example used in the book chapter is whether there's a link between video-gaming and aggression because there's quite a lot of research that seems to like to try to suggest that the more video games you play or violent video games especially the more aggressive you may behave so there's there's a link to kind of externalizing disorders and combat problems so you might have a situation where you basically have a relationship a well-established relationship between say video game use which you've meant be measured on a continuum so how many hours a week that someone plays and an outcome or say aggressive behavior now moderation would be where we're talking about some other variable changing that relationship so the relationship between video game use and aggressive behavior is changed in some way shape or form by some other variable something's on to the book in the book the other variable we had are callous and unemotional traits so this is a dispositional kind of tendency to basically not be a very nice person and that's also been linked to conduct problems and externalizing problems so it kind of makes sense that there might be some kind of kind of complex relationship going on between conduct problems or aggressive behavior video game use and these callous unemotional traits now what moderation implies is if you if you look at kind of what it means theoretically you've got some kind of predictor in our case that's going to be video game use some kind of outcome variable which is going to in our case the aggression and the moderator these callous unemotional traits they moderate this relationship so we've got this kind of arrow between our predictor and our outcome so that's a you know literally kind of you can imagine that sort of a correlation or a beta in a regression model so there's a quantified relationship between video games and aggression and what we're saying is the the size or the nature of that relationship changes as a function of callous and unemotional traits so there's a few ways that this could work now first of all it's possibly easiest to explain if you imagine that callous unemotional traits is a categorical variable so you either have them you know you're a callous person or you don't have them you're not a callous person so if we were to collect data that number of hours playing video games aggressive tendencies and then we classify people as having these callous unemotional traits or not so therefore one or two groups then we might see something like this graph so what this shows us is that the relationship between video games and aggression playing video games and aggression for non callous people is actually pretty much a flat line so it's the red line in this graph here so it's this this is not going to make you much clearer on there you go if that is that red line there so it's a flat line there's no relationship there that's all pretty much so doesn't matter how many hours you play video games if you don't have callous unemotional traits you don't and you don't show greater signs of aggression however for our group who do have callous unemotional traits we see this blue line here I'm drawing this with a mouse which is not the easiest thing the world to do I'm doing it left-handed when I'm right-handed don't ask why and so for our callus group there is a positive relationship so we get this kind of blue lines showing a very distinctive positive relationship so this is an example of moderation so the relationship between playing video games and aggression is different it changes depending on whether you're callous or not callous so that's what moderation is it's kind of saying the relationship between two variables so in this case differs across the two groups and one group is it's virtually zero in the other group it's kind of positive but it changes in some way and the more complex way to look at this is if we imagine that callous unemotional traits are measured on a continuum and we can display this graphically like this so I should say this diagram took me a whole day to create which not sure what that says but anyway um so we've got a situation on the left hand side of the screen so this one over here still using the left hand there of no moderation so basically it says it's situation where callous unemotional traits do not moderate the relationship between video games and aggression so you can see we've got aggression as our outcome we've got video game to the predictor and the nature of the relationship is shown by the slope of this plane this from this regression plane and so if we look at the bottom in so this is when callous unemotional traits are at a very low level so remember they're measured on a continuum and you can see this blue arrow represents basically the relationship between video gaming and aggression so it's a slightly positive relationship now as we move along the scale of callous unemotional traits we could say well what about someone's got kind of a medium amount of cattle an average amount of callous unemotional traits what's the relationship between video gaming and predicted aggression there and again I've drawn on a blue line to represent this and you can see it's a pretty similar slope to the the blue arrow when we were looking at low levels of callous unemotional traits so in other words the slope of this regression plane isn't isn't changing as callous unemotional traits increase we don't see a change in the relationship between video gaming and predicted aggression now what happens when we look at the top end in the scales this is high levels of callous unemotional traits again I've got blue arrow that that shows and a relationship between video gaming and aggression so it's just running parallel to the regression planes and just shows you kind of the slope of that plane and again there's not really very much of a difference so essentially there's the slopes of these blue arrows here do not change so as callous unemotional traits increase we don't see a change in the basic relationships in video gaming and aggression now what we have on the other side of it on the right hand side is an example where we do get moderation so what we're just talking about was when callous unemotional trace isn't isn't a moderator when it is a moderator we get mmm I mean well this is an example of the sort situation we might get although it's a you know it's a bit over eggs or whatever so again if we look at low levels of callous unemotional traits then the relationship between video-gaming aggression is actually negative so this blue arrow is pointing downwards indicating that the more video games you play the less aggression you show at low levels of calais traits what about medium levels of calais traits now so we're looking kind of here again we've got another blue arrow indicating that the sort of shape of the relationship between video gaming and aggression and this hour is pretty flat so it's kind of showing that there is a relationship between video gaming and aggression what about our high levels of callous unemotional trait so right at the top end of the scale again we've got blue arrow to show us what's going on and here we start getting a positive relationship so this arrow pointing upwards so that's indicating that as video the more video games you play the more aggression your you display so this is an example of moderation in that low level of callous traits there's actually negative relationships between video and gaming and aggression I kind of average levels of Commerce traits there's basically no relationship between video gaming and aggression and at high levels of callous unemotional traits there is a positive relationship between video gaming and aggression so the relationship between video gaming and aggression is changing as we move along the continuum of callous unemotional traits so that's what moderation is in terms of how we test this statistically we basically do a fancy regression so we do a regression in which we put the predictor the moderator variable and most importantly the interaction between the two and it's this interaction that tells us whether we have moderation so if that interaction is significant then it means that the moderator is moderating the relationship between the predictor and the outcome so statistically how we would test our video gaming example we'd have our outcome of aggression a predictor would be outspent video gaming moderator would be where the person lies on this continuum of calais traits and the we would put in an interaction term which tests whether basically the relationship between predict when the outcome changes as a function of the moderator so that's moderation and if you literally wanted to write out the aggression oh sorry the regression model it would look something like this so you predict aggression from gaming callused rates and the interaction and often to here's the data for our video game example so you can see that we've just as ever with each participant is a row in the SPSS data editor and each variable is a column so we've got an ID variable just you know identifies who the person was then we've got a score on the aggression variable we've got a score on the video game variable so this was hours per week that they played video games and we've got a score on the callous unemotional traits tells us we're on that continuum the person fell so to run a a moderation model is really pretty simple so now we've installed these lists these Andrew Hayes dialog boxes we can just go to analyze and regression and then select the process tool and what we get is a dialog box hmm which is pretty straightforward and now you can fit many many different models so here whether you get a drop-down box model number this process to will run 74 mmm different types of moderation mediation type of analyses of varying complexity and the ones that I cover in the book are models one on form one is a basic moderation model and four is a basic mediation model so the default option here is four because lots people will use this for mediation but we want to change it to one because we're running a moderation model now it's really very straightforward so we've got a slot for our outcome variable so our outcome was aggression so we can pop that over there and our independent variable that's our predictor so our predictor was the hours spent playing video games and then the end variables are any moderators that we want to put in now we've only got one moderator you could put several in and you know do more complicated models and this and and if you want to do that read Andrew Hayes's book is the best advice we've just got it's one more race so we can drag it across like that and you know in a sense that's all there is to it now there are some other options which are going to be more detail in the book about the important ones which I won't really kind of explain are you follow up moderation generally with a simple slopes analysis and if you want to plot the results of that then it's useful to take on this box generate data for plotting and it's good to select heteroscedasticity consistent stand of errors because if you've got heteroscedasticity that just means that your your standard errors are going to be okay so it's a pretty good thing to take anyway and the other thing from reasons again explained in the book we quite often do something called mean a grand mean centering when we do moderation analysis and basically you can get process to do that for you automatically by ticking that box there so these are all useful things to tick suffice to say so select them and click on continue and again it's quite useful to have a look at or basically to follow up this analysis by asking for various things that sort of break down the moderation effects and you can get those through this conditioning dialog box here so just select both of those I'm not going to go into a lot of detail but again it is explained in the book so click on continue that's all there is to it click on OK and hopefully we should get some output at some point mmm couple things worth noting while SPSS is doing its business is you've got to be a bit careful with variable names so what the process tool does is it converts variables to two eight character strings so if you've got very that are the same for the first eight characters so say you had like aggression one and aggression to and you put them in the same analysis then process will get a bit confused by that because it would abbreviate them both to to aggress e it will squish them to eight characters and because they're the same and our characters long you'll get an error message so we can see for example in this output that's this is what it's done so for each variable it's abbreviated it's at eight characters so this was a really cool digression and it's been abbreviated to aggress e and did games was a has been abbreviated to vid game because again it's been cut about eight character so just be aware of that now in terms of interpreting whether we have significant moderation or not it really couldn't be simpler than just looking at this part of the output down here where we basically have a regression so it will have you know centred things for us and we so there's a basic regression table so we've got each predictor callous unemotional trades video games and int one is the interaction and it tells you below that this interaction represents video games by callous unemotional trades so what we're really interested in doing in terms of interpreting weather moderation has occurred is literally looking at this p-value for the interaction and if that p-value is less than point zero five is the convention then we say there's been significant moderation so we can see callous unemotional trades very significant predictor of aggression video game use significant picture of regression but the most important thing in terms of moderation is this interaction term which is significant and also this confidence interval does not include zero so this beta value is in the in the population is likely to be more than zero so in other words there is an effect in there and moderation effects in the population in terms of how we followed this up we use simple slopes analysis which as I said I'm not going to go into in massive amounts of detail particularly all I will say is we can look at this part of the output here these are the conditional effects of the predictor on the outcome so this is this first row tells you whether we're looking at and you can kind of ignore the numbers really this is solved low levels of callous unemotional trades zero is average levels of callous unemotional traits because remember it's in centered so zero is is them is the mean because it's been sent in around zero and this third value is basically a positive or like a high it's these are actually kind of one one standard deviation bubble below the mean so this is kind of like some deviation below the mean the mean level and standard deviation above the mean but you can think of them as just sort of low average high if you like and then the effect is basically again another regression coefficient so this is the the effect of video games on aggression at low levels of callous unemotional traits and if you look at the p-value its non significant so at low levels at callous unemotional traits you don't get a relationship between video gaming and aggression there are medium levels of callous unemotional traits again if you read the peeve the beta there is a significant effect and it's a positive beta so that's showing as a positive relationship between video gaming and aggression and finally a high levels of callous unemotional traits the beta value has gone up it's it's a more positive relationship and that's highly highly significant so basically as we move through the continuum of callous unemotional traits the relationship between video gaming and aggression is goes from non significant to positive and significant to even more positive and even more significant and that's at a basic level that's all there is to it so what about mediation well in the book chapter we've got a different example for mediation and this is based on some real research actually and a real research example so the data are like you know there's someone's actual data and this was a study looking at um the relationship between pornography consumption and infidelity though is a study by Lambert Excel and what they were interested in is whether this relationship between pornography consumption and infidelity was mediated by relationship commitment now mediation is a different kettle of fish from moderation so basically you've got you start off with some kind of relationship so in this case we will be predicting that pornography consumption and has a relationship with infidelity so that the more pornography you watch the more likely you are to be unfaithful now what mediation is all about is is whether this simple relationship operates via a third variable so in other words is it the case that pornography consumption has an effect on infidelity because it's influencing some other variable so the mediated relationship looks we get this off triangle or variables so we have our predictor pornography consumption and it's predicting some kind of outcome which is infidelity and what we're suggesting is that there's some factor that's related to both of them that explains this relationship in some way so in this study what they predicted it was was relationship commitment so they were basically saying that the more pornography you watch so there's the as the predictor changes that has a significant effect so that's this red arrow here has a significant effect on your relationship commitment so you start feeling kind of less committed to your partner and relationship commitments or separately has an influence on whether you're likely to be unfaithful or not so this red arrow here pathway B indicates that so we've got pornography affecting your relationship commitment we've got relationship commitment affecting your likelihood of you being unfaithful and what mediation is suggesting you know or if you have mediation is it saying that this relationship between your predictor and outcome will be weakened in ideally if it's perfect mediation it will be reduced to zero by including this mediator so the relationship we've called it see here between the predictor and the outcome once you include the moderator sorry the mediator in the model becomes zero so we'd be saying this or C - it's called the direct effect is reduced to zero so if we have mediation what it means is we should get a very small direct effect between our predictor and outcome but a very large indirect effect so the indirect effect is is the effect of in our example the effects of pornography on infidelity operating through relationship commitment so that would be the indirect effect whereas the direct effect is well as the name suggests it's the direct influence that pornography consumption has on infidelity so we can see the actual example here all I've done is just slop the names of the of the constructs into the boxes and so mediation would be shown by basically a significant indirect effect so we can quantify this indirect effect we can quantify and basically the the combined effect of these pathways a and B and if that is significant then we we basically have it as a significant mediation so you know in other words if the indirect effect is is kind of big enough then our relationship between pornography consumption infidelity has been mediated by relationship commitment now historically there are kind other ways to think about mediation and this comes back to this idea of this direct effect being close to zero or being very much reduced and compared to when we didn't put our mediator into the model but in terms of how we look at in the book we talk more about quantifying this indirect effect and seeing whether it's significant or not doing mediation you can see here we've got the data file from the book for the land ASL study and three variables we have we have pornography consumption actually going to use the log transformed values where we've got a variable commitment which is relationship commitment scored one to five and we've got a variable representing infidelity scores from zero to three so as ever in SPSS each row represents a person in this study and each of these columns is representing a variable the variables were interested in they're just these on slow consumption commitment and infidelity so to do mediation we again use Hayes's process tool so it goes the analyze menu in the regression menu we'll have this process tool that we've now added in and you can see we get this dialog box it's the same as the one we use for moderation this time we want because we want to do mediation we use model number four that's the default so we don't need to change that we've got an outcome variable in this case our outcome variable was infidelity so we can pop that over there our independent variable that's the predictor so in this case that was pornography consumption and the end variable this time is gonna be our mediator and a mediator was relationship commitment so we can pop over there so basically that's kind of all we need to do and it will do bootstrapping for the indirect effect and the default options for that bootstrapping is fine and we can have a look at some of the options this gives a slightly different dialog box the first some of the options we looked at for moderation don't apply for mediation models so for mediation models we really want to look at the four options down the bottom we can ask for some effect sizes I'm not going to go into them but they're explained in the book if you want them you may be familiar with a serval test which is another test that's quite often used in mediation personally I just look at the indirect effect and not bother with the Sobel test but you can select it if you want we can ask it for some information about the total effects model and that can be useful and we can ask it to compare indirect effects which also can be useful so you can maybe select those options click on continue and then if we click on OK it will merrily do a mediation analysis for us hopefully because using bootstrapping this can take a bit of a bit of time to do and so I'll just sit here awkwardly twiddling my thumbs waiting something's happen and then eventually twiddle twiddle twiddle twiddle tool so it's running through all there we go so we've got some output so what this output tells us now first of all we get a kind of well we basically get a regression for pornography consumption as a predictor of relationship commitment so this is just checking that our predictor variable predicts our mediator and we've got a regression coefficient for it and you need to sort of check the p-value to see whether that's less than 0.05 in this case at ease which means that pornography consumption significantly predicts relationship commitment which is you know useful and now the second block of text down here is referring to a model in which relationship commitment and pornography consumption are both predicting the outcome which was infidelity so we can have a look whether there where the relationship commitment predicts infidelity it's got beta of minus 27 that's highly significant p-values as 0 to 4 decimal places so that's very significant predictor and again we can look at pornography consumption as a predictor of infidelity and that is still that's significant with obese or 0.46 significance valuation point oh five so yeah that just sort of tells us that the relationships you would expect to be significant are so both relationship commitment and choreography predicts the outcome which is important but also it's important that pornography consumption predicts relationship commitment which we saw in the first model so then in terms of working out whether mediation has happened the main thing that we're interested in is here this indirect effect of X on Y so this is the end of indirect effect of pornography consumption on infidelity and it's the indirect effect via commitment so this is why it's labeled committing there its commitment to eight characters so the question is is this significant and the way to tell whether it's significant is that's the size of the effect so that's kind of a B so valued for the indirect effect point one two seven and we're interested in whether this bootstrapped confidence interval includes zero or not so this confidence interval is going to give us a rough idea of what the population value of that indirect effect is it's a 95 percent confidence interval so if it doesn't contain zero then basically our effect is significant at the 0.05 level but more important it means that the population value is not zero or is very unlikely to be zero anyway so we've got lower confidence interval here of point zero one seven and we've got an upper upper value of conference in twelve point two nine six so that means the population value of the indirect effect lies somewhere between point zero two roughly at about point three so in other words it's positive so that confidence interval doesn't contain zero so the population value is is likely to be a bigger value than zero in other words that is an indirect effect of commitment so in terms of determining whether you have a mediation effect that what you really need to do is look at this indirect effect and whether the confidence it will contain zero if it doesn't contain zero then you can conclude that you have a significant mediating effect the rest of the output mainly is to do it effect sizes which I'm not going to talk about that if you're interested in those then get you know still steal the book and have a look because that does talk about these sorry output in more detail okay and that's all there is to it

Channel: Andy Field

Views: 378,951

Rating: 4.9213438 out of 5

Keywords: Moderation, Mediation, SPSS, Statistics, PROCESS tool

Id: RqkGMqDU20Q

Length: 34min 22sec (2062 seconds)

Published: Thu Mar 07 2013