Moderation analysis in SPSS using the PROCESS macro

today's video is going to be focusing on how to undertake a moderation analysis in spss so what is a moderation analysis or moderator analysis it is an analysis used to determine whether the relationship between two variables x and y depends or is moderated by the value of a third variable so what this means is that x and y already have a relationship and you could assume that it's statistically significant so x would predict y in a normal linear regression however to create it i suppose to to develop a model that explains a greater amount of variance you hypothesize that there is a third variable which essentially as that variable score changes the relationship between x and y would change as well so there are some assumptions for a moderation analysis and these assumptions are the same that are required for a linear regression say linear regress there but it means linear regression so the assumptions are that the dependent variable is measured on a continuous scale there is independence of observations data must be approximately normal there needs to be a linear relationship between the independent variable the dependent variable and w which is the moderator variable the data needs to show homoscedasticity and the data must not show multicollinearity so there's there's a level here so they need to have a linear relationship but they can't be so correlated that essentially they're measuring the same outcome generally the the limit is said to be about a correlation of 0.5 so if it exceeds 0.5 you're you're starting to edge along the lines of multicollinearity which should be avoided and then the last two assumptions are no significant outliers and residuals are approximately normally distributed i go over how to test these assumptions in some of my previous videos which i'll hopefully link up as i'm speaking about this and if not just check the history they're pretty easily titled all right so the conceptual model of a moderation analysis looks something like this on the the left hand side so we have x our independent variable and it has a direct relationship with y our dependent variable and w here is our moderation variable our moderator variable which has an influence on the relationship between x and y so that's how the conceptualization of that model would look something like this and the statistical model would look something like this so you've got to understand firstly that a moderation analysis is essentially a linear regression and in fact you can do moderation analysis using essentially a linear regression model so in in that case if you're using a linear regression model you would simply get the product of the independent variable and the moderator variable so x times w and then add it as suppose as a next block in a in a hierarchical regression model and determine if that interaction effect that product of an independent variable and a moderated variable is statistically significant and if it is then you would have proof that a moderation effect exists between those two variables on the relationship to y right however i'm not going to be providing an example of that model today today i'm going to be showing you how to use process which is a macro developed by this chap called andrew hayes which is extremely useful and it has the added benefit of if you have a significant model of giving you syntax to plot out that interaction effect so you can see exactly where those changes occur and at what levels which is i find very helpful in interpreting the outcome as opposed to getting a positive or significant interaction effect which then you'll need to use linear equations to determine the difference stages of um the interaction so to avoid that let's just use andrew hayes's macro called um process right so if you go into your web browser hopefully not internet explorer but if you do no judgement and go to [Music] processmacro.org and you will come to this page here which is the process macro for spss and sas we're going to focus on spss here so what you want to do is go to the download tab and then scroll down here of course you got to read through the copyright and disclaimer and then download process so download the zip folder you save that wait for it to download hopefully you internet than i do right so once you have downloaded your zip folder just simply open it and you can extract it to wherever it's most suitable so something some a directory that will stick around so you're not going to be deleting anytime soon i'm going to put on my desktop for now but um i'd recommend putting somewhere i suppose more long term so just ex just extract all the information inside there to your process folder wherever it may be and that's what you have inside so there are some quite useful documents inside here so little guide syntax the updates on process history and some other bits and bobs right so what you want to do is then open up spss if you haven't done so already and then you want to go to extensions utilities and install custom or install custom dialogue right so once you've clicked on that then you want to go to wherever you downloaded or wherever you download the process macro to and extracted that for my case will be on my desktop then you go to spss custom dialog boulder file and then click on process dot spd right if all goes well it should say the dialogue files and stored to xyz and you're good to go right so then you'll be able to see this macro menu here if you go to analyze regression and process so i previously had process 3.1 and it's now been updated to process 3.3 i'm not sure what has been included or what has been changed but i'm going to assume that not much as so at this stage you now have the macro installed and you can proceed with analysis so i'm going to open the data sets this one is going to be called office dot sav so essentially this is data set with around 427 individuals so we have the id number the age of the participants continuous variable support which is a measure of support available from colleagues a scale or continuous variable then the next variable is work demand so how often do you meet work problems and then stress which is essentially a measure of stress now if we go back to the model the conceptual model i'm going to be testing whether we go back to the conceptual model i'm going to be testing whether work demand relates to stress so does work demand predicts stress that's our first hypothesis or first test but i also want to see if the support that employees receive moderates that level so if employees receive a high number or high amount of support do they have less stress regardless of the fact that they have a high workload or work demand and if they receive little support does that make the effect that work demand has on stress even stronger so that's the model that i'm going to be testing so it's a simple moderation model all right so let's get that out the way so so that's so that's the model we're going to be testing and i also want to add age as a covariate so i want to control for the influence that age might have on any of these relationships because you might argue that people who are younger might need more support or perhaps people who are older need more support and they get affected by work demands at different levels so we want to control for the effect that age might have on confounding the relationships that we're investigating okay so i'm going go to analyze right so before we get started us we're going to make the assumption that all of the assumptions that are required for linear aggression have been met for this data set so the normal observations i mean so they are independently observed they are approximately normally distributed they have linear relationships and so on so when you're doing your own analysis you're going to make sure that these assumptions have been checked off and you're you're confident that they meet the assumptions required for the analysis otherwise the results you might get may not be what your data is actually telling you right so we're going to analyze regression and let's try use 3.3 hopefully nothing else has changed too much so here we get our little box for process our y variable which is our dependent variable in this case it would be stress because we are trying to predict stress and we think that work demand work demand is a predictor of stress that age is a covariate or something we need to control for and support available from colleagues moderates this relationship between x and y so the assumption is you always use w when you have one moderator so w is your first moderator and z would be your second moderator if you have two moderators in the same model not going to be covering that today but just so you know also another caveat you need to make sure for easy interpretability that your variable names are less than eight characters or eight or less characters long because if they're greater than eight characters then process will cut them out and if you have a long variable within similar stems you're going to be quite confused as to what variable has been um reported so try to keep them to eight or less characters right so once we've entered our variables into the correct places we're gonna go to options and there's several things you can you can choose here but i'm really only interested in the generate code for visualizing interaction so that's going to be limited to that for this video and just a a quick um just a quick mention here of conditioning scale so you can either choose the 16th 50th or 84th percentiles to have comparisons so we'll take the moderator at each of these levels the moderator scored each of these levels and then look at the effects that it might have had on x and y at these different levels so essentially i would stick with the 16th 50th and 84th percentiles purely because if your interaction effect is not normal so your variables need to be normally distributed but if your interaction effect is not normal then it's a more accurate representation of differences at the lower medium and high end of that score and if it is normally distributed then the 16th percentile is equivalent to what -1 standard deviation and the 50th percentile is equivalent to the mean and the 84th percentile is equivalent to plus one standard deviation so essentially if it's not normal if it's not a normally distributed interaction you need to use this and if it is a normally distributed interaction then using this is the same as using the standard deviation so my recommendation would be to just stick with the 16th 50th and 84th percentile right so we've got that sorted and we're going to click continue we don't have any multi categorical variables and we're just going to click ok and here's process running its macro i think the code itself is around 4 000 lines so it takes a little bit of time to get done and here we get our first output so here we have model one ah actually let me go back i forgot to mention which is an important point so we have here we have model numbers so there's a lot of model numbers here and each model represents a different conceptual and statistical model that i presented earlier so in process model one is the conceptual model presented here so would have one moderated variable on the relationship between one independent variable and one dependent variable so you want to make sure that your the model number you choose is corresponds to the model that you want to test so andrew hayes used to give out these model the conceptual models and statistical models for each of the model numbers and explain what they're testing but he doesn't now so you'll need to get the second edition of his book which i'll add into the link if you want to have an overview of what each model means and if not then you can watch these videos as i slowly progress through the different models all right so just make note of that model number is important in process so we've already got the output so i'm not going to run it again so here we go model one so we have a moderation model we have a dependent variable of stress a moderation variable of work i mean an independent variable of work demand and a moderation variable of support the total sample size is 427 and over here we have essentially what is a regression model so overall our model is significant 0.000 goes through the room we have an r squared of approximately fifteen point six percent so our model explains fifteen point six seven percent of um what stress is comprised of in this data set and here we have our normal regression output so we have work demands we have a this is essentially unstandardized coefficient so beta score so work demand is a significant predictor support is not and age by itself has an influence on stress so that's essentially that's what we're testing so what what effect do these models and unison have as predicting stress but as we're looking at an interaction effect moderation effect the individual p-values and coefficients of these variables falls by the wayside we're not particularly important interested in them they're not particularly important by themselves what we are interested in however is the interaction effect we can see here this is int1 and this is the interaction between work demands and support on stress and we can see here that this interaction is significant 0.0366 so that means that our interaction effect is significant and here we can go down to see so support here remember we chose the 16th 50th and 84th percentiles so these these support scores here on the left hand side represent the 16th 50th and 84th percentile of support so that would be the average score in that percentile we can see the effect that it has here now if we want to visualize what this interaction effect means we have the opportunity to use this little syntax provided to us by andrew hayes very useful so what we would do is we would highlight and copy this go to a new syntax paste it and highlight it again and then run it so what this creates is a new little data set just with those variables and those figures and allows us to have to visualize what's going on in that interaction so here's the basic outputs just want to quickly interpolate the point so it's easier to see so you go to click on the dots and then go to elements and click the interpolation line now we have some actual lines so we can see how it changes over time so again here's the support which is the moderator variable and then we have blue which represents low support red which represents medium level so 50th percentile and green which represents high level of support and on the y-axis we have stress and on the x-axis we have work demands see how we can see how support changes the interaction between work demands and stress so let's assume on the first points that this is relatively low work demand here we can see that individuals with high levels of support have low levels of stress than people with medium levels of support and people with low levels of sport even though there's low levels of work demands have substantially higher levels of stress than all other percentiles or all the levels that we're looking at with regards to um levels of support attained or given now as work demands increase for the people with high support we can see that the overall levels of stress doesn't change pretty much at all so that means that support mitigates the effect that increased work demands would have on the individual stress level and as as support tends to starts to drop down for these individuals the effects of work demand on stress start to increase so for the 50th percentile it slowly starts to creep up in a linear linear relationship as work demands increases and individuals with low support so the 16th percentile of support there's a it's quite a substantial increase from low levels of work demand to higher levels of work to mind on stress so that is essentially what an interaction effect is like when it's illustrated and it also helps to be able to interpret what that effect is telling you in a visual format as opposed to just looking at some output and that is one of the prime reasons that i think process is a great tool as opposed to using the linear regression model because essentially process is a linear regression model but with some extras with some extras added in and yeah so that's that video i hope it was useful and let me know if you want me to go over different models and process such as a mediation model moderated mediation and so on i will be happy to oblige

Channel: PsychED

Views: 45,566

Rating: 4.9642859 out of 5

Keywords: PROCESS macro SPSS, moderation analysis, moderation SPSS, SPSS statistics, SPSS tutorials, andrew hayes moderation, how to SPSS, XW moderation analysis, moderation linear regression, moderation mediation

Id: p0UbBJwFoeA

Length: 21min 33sec (1293 seconds)

Published: Sat Jun 29 2019