SPSS - Three Way Moderation - Interactions with PROCESS

so in this video today I'm gonna go over how to use the process plug-in in SPSS by Andrew Hayes to do what I would call it a moderated meet moderated moderation model or 3-way interactions and so this is model three in the process plug-in which you can see in the picture here so let me explain to you kind of what's going on I don't have the best example for this demo but I could show you how to run it on your own data and so I have a data set that has Student Success as the dependent variable so if we're trying to measure engagement and GPA and like a combined sort of flank are they going to be able to matriculate for the University so we've got a Student Success variable for X here we have an AC t score but we've used it as a cut-off score so there are specific cutoff scores that at least our university uses to allow students into a regular version of an English class or math class if they make below a certain score they have to take a remedial class first so we're kind of seeing if that a CT score actually predicts the success in this course and so is it a good criterion to use our moderator variables include student GPA so high school GPA but it's scaled so it won't be in your normal GPA and scale and then ours studied according to the student it so we're going to use both of those as moderator variables but even though you can put multiple things in M in the process window you want to make sure you match it to the model that you're using so we're going to use M and W and so what this picture denotes is that there is a moderation of M on the relationship between x and y so there's an interaction between X and M to predict Y but there's also W and so WL moderates the relationship between M and X and y and so essentially what that amounts to is a 3 weigh-in action between xwm and that all the two-way interactions so X and W X and M and then W M if you want to look at the statistical diagram you can see that a little better so we have our three main effects of M X and W on y and then I have my two interactions with M my one interaction with W and three-way interaction so this is essentially a three-way interaction with the process plug-in helping us out all right so we're gonna use model three and here's the data set and so all of this table eventually be available on our design site we're designing right now stats tools calm and you'll be able to practice on this actual data set but for right now if you need the data sets hopefully in a couple weeks this part of the video will be useless you can send me an email my email is Erin Buchanan at Missouri State DOT edu or Aggie Erin AGG ier I am at gmail.com or you can just contact me through youtube that works too alright so getting into this data I first want to do my data screening because it's important to look for outliers and I already know I have some in this data set and we'll have to decide what to do with them so if you've watched some of my other videos you have a very specific order for data cleaning so we're going to start by checking for missing data so did an SPSS I'm going to analyze descriptives and then frequencies you move all of your variables over so let me reset this so I'm going to move all of them over imma turn off frequency tables because these are continuous variables and it'll give me a ton of output if I don't and it warns me that unless I do something else I am wasting my time so let's click statistics and then at least ask for the min and the max and that will give me uh any out of range scores as well I already know this data is within reach it's submit continue and okay it's the first thing that does is it gives me a box that tells me what I have a or my minimum max so these scores don't necessarily match what you'd expect for GPA because we scaled it a certain way and then hours attended some study sessions I have their six actual success score Emma a CT score is either below or above the cutoff I have some missing data here so I have to and the GPA too and the success in the core or one of the success in the course and then 24 in this hour studied now I probably don't want to fill that variable in because that's ten percent of the data so I want to leave that out now though for scale GPA I could probably fill that in because it's only two people and then for success in the course I tend to avoid filling in the dependent variable because that's what you're trying to predict and so you're guessing at a an average score easily and but you could because it's only one data point it really probably won't affect a whole lot but for this data set I'm just going to fill in my two missing for skilled GPA I'm gonna leave everything else alone and then SPSS will ignore those because they're missing so you can do that fairly easily under transform and then replace missing values the best option in SPSS is linear trend at point which you need to change first so otherwise it will automatically do series mean so often you can move stuff to the right and then um click on what you're wanting to include but this is one of the weird windows where you need to change it first so let's do a linear trend and I just want to do GPA and you'll see here that's this trend so that's how I know that it changed okay all right so I'm actually going to get rid of so let me save this as a new file real quick I'm actually may get rid of my first GPA column so I'm not tempted to use it now when you're doing this sort of thing I always tell people to say that as multiple data sets so you have that original data somewhere else but and then get rid of the columns that you aren't using so you're not tempted to fill in missing data and then analyze the non fill vent column that happens a lot all right so we've gone through missing data and dealt with that everything else we're going to leave missing pairwise let's deal with outliers now we're actually gonna run this analysis with multiple interactions but what I want to do is screen the data before we create these interactions you could also create the interactions and screen those but generally since those interactions are part of the main effects people who have outliers and main effects aren't going to suddenly go away because of the interactions and so to do that since this is a real regression we can sort of run a non interaction regression to analyze that data so let's go analyze regression and then linear because you can't do this sort of thing in the process window so what I want to do is use my success in the course as my dependent variable because that's going to be my dependent variable in the analysis I'm going to put all three of my IVs in the independent variable box as I am going to use those to predict as main effects under plots I'm going to do Z pred and Y Z residual and X that will give me my histograms and my residual scatter plots and so I want to click on both of these a my normal probability plot for linearity it's a histogram a normal probability plot that continue under save we're going to click Mahalanobis cooks and leverage to check for outliers continue and okay so first thing I did was actually give me the regression so I could kind of see like what my regression is going to look like but that's kind of less cheating so I'm going to go back and look at my Mahalanobis scores here so I'm going to sort those descending so right click sort descending and that'll put the biggest ones at the top well how do I know what's wrong like what scores are bad so I have a chi-square table here I remember that you want to use point zero one so we want to find scores that are very odd before we start deleting anybody and so let's see so for three a cut-off score of three how did I get three well I have three independent variables so those you want to use a degrees of freedom with a number of K variables and so that's three I've got what XW and M and then I'm going to use point zero one so sixteen point two seven will be my cutoff score from Mahalanobis so I have three Mahalanobis outliers and then I'm going to use the best to a three rule and make sure I only delete people who are outliers in multiple ways in a regression because often people can be discrepant they can be far away from everybody else but not have any leverage which means control over the slope so the people that you really don't like are the ones who have big control over the they move the slope up and down because we want our slope to be fairly consistent so I'm going to use all three of these and look at if a person has a outlier on at least two out of the three and then I'll think about deleting them so how can you keep track of that because it's a lot of scores at once we're going to use the recode function in the transform window to help us do that so I'm going to recode in different variables so transform recode into different variables kind of see what's going on here in a second so take my Mahalanobis score and I'm going to call this Mahal out and what that will do is create me a variable that is simply a 0 or a 1 that tells me if that person is an outlier or not on that one score it's a random through this three times because you can't use the same cutoff score for all three columns but then take that long a sort of my old and new values and so what I want to do is range value through the highest so I'm going to take my cutoff score which is sixteen point two seven and I want to take everybody whose score is sixteen twenty seven or higher I want to make them a one for yes and outlier hit add and then all other value I'm going to make zeros no they're not an outlier and then add I continue and ok and that'll give me a new column with my three people marked as Mahalanobis outliers and then everything else will be zeros for not being metal R so for cooks here so let me writes sort descending for cooks these are all positive that makes it nice and easy so cooks is a measure of discrepancy and leverage together is a remember discrepancy is people who are far away from the data so there's a big gap between them and everyone else and leverage is how much control they have over the slope and so the cutoff score of cooks is um and I'll type this out so you can see it the whole thing bigger yay okay so it's four divided by degrees of freedom which is n minus K minus one so in this particular example that's gonna be four divided by two hundred sixty-five participants minus three for K minus one and my math skills not so hot when you're talking about small decimals so for know I brought up that particular picture it normally does a calculator for you there we go so 265 minus three minus one it's 261 so 4 divided by 261 is 0.01 five and let's go for decimals oh one five three and then we're gonna do the same thing again so transform um recode into different variables my reset this so let's take Cooke's and make it the Kochs outlier click older new values it's still ranged through highest because they're all positive they're going to go oh one five three and make those people once and and then all other values and make those people zeros and then add so continue and okay so I have quite a few cooks outliers so that is 20 people who are a combination of discrepant or leverage and so I've got them on marked but so far it looks like that low overlap is only this one person who's not wire on both and then last but not least let's do leverage so the cutoff score for leverage is 2k remembering that K is the number of predictors divided by 2 k plus 2 divided by n so that's 2 times 3 plus 2 is 8 divided by 265 go back to my calculator here so I got 2 times 3 plus 2 divided by 265 so that point Oh 302 5 round up the cutoff score is 0.03 0-2 let's do this one more time so transform recode into different variables and let's hit reset I got leveraged values so let's do my leverage outlier change get old and new values range value through highest that's point O 3 o 2 and the value is going to be 1 hit add and continue oops just getting old new values now let's also do all of the values are zeros I want to leave it like you continue and okay I forgot to sort leverage here so let's sort that bad boy descending so I have a lot of leverage people as well so 30 participants who have a high leverage so from there what I want to do is pick the bed the worst two out of three well that's still kind of hard to see it's pretty clear some of these ones at the top or two out of three what which ones are they so what you do the easiest thing to do is transform and compute variable it's a lot of the transform window we're going to do an outlier sum and I'm going to add these three columns together so Mahalanobis outliers plus Kooks outliers plus leveraged outliers hit okay and then let's sort that descending and now it's I can see that I have 13 participants who have at least two out of the three are bad and these are the participants that I would want to exclude in my analysis if I decide to exclude outliers let's do that you don't have to which can so what I'm going to do is save myself a new data set and I would normally call these something better than one two and three but I'm just using this for the video example okay and I'm gonna get rid of these people 13 participants deleted due to outliers and when you write this up you would explain they used Mahalanobis Kooks and leverage as your criterion for deleting those people all right mmm so now that we've dealt with missing data and outliers the next thing you would want to do is multicollinearity so you don't want to waste your time using 2x variables that are very highly correlated and especially not when doing moderation because then when you add the interaction the multicollinearity gets to be too much okay and they might suppress each other will read my net run so let's do analyze correlate bivariate just your independent variables so you got a bunch of extra columns here but just your three independent variables so I got my GPA my sessions and my a CT score okay and the a CT score one will be a point biserial because it's only two variables and we will leave the DB out you want them to be correlated with a DV that's the point of regression so we're not going to look at that correlation because that's what we're trying to do I'm hit okay go back to my output here and it looks like um we're not at multicollinearity levels yet but it's pretty high the correlation between their GPA and there are hours of attended sessions or study sessions is pretty high so as GPA scaled GPA is going up and they're going to more sessions and that shouldn't be too surprising because the smartest students tend to know that they need to study more lots of research showing that poor students don't know that they should when to get help until it's too late and so you often find that any time you have optional study sessions students who need to go don't but there doesn't seem to be a relationship between GPA and a CT score which is a little odd but that's partially because we've got this scale is a categorical variable and so there is a correlation and it's negative so higher GPA the lower a CT or the more likely they are to be in the 0 group which isn't necessarily the best correlation to have which is why I said I don't have a fantastic example for this particular video because that doesn't make a whole lot of sense all right but anyway do we have most collinearity no and that's good okay I might worry a little bit about this point six because that's 36 to 40 percent of the variance is starting to overlap which is kind of a lot so we'll see what happens the last piece is the exceptions and we've actually already run them up here and they're going to be kind of scary but I deleted outliers so I want to rerun it to look at assumptions again so analyze regression and linear you can just leave the whole thing set up you will get extra Mahalanobis and leverage columns which you can just ignore them I'm hit okay and let's look at our assumptions so do we meet multivariate normality whoa that's kind of iffy I'm one problem that you have is that most of the data generally is between two and two on a scaled residual um and then you have all these like dudes out here so most of it it is actually centered over zero which is what you'd want and you've just got a couple of little residuals out here that are really bad out um outside of the data because we've deleted outliers but they the residuals are pretty high so you might take some time to figure out who those people are on why they're so high but I'd say this is probably a cage given the fact that most of the data is here and normality is usually considered robust to violations with at least 30 participants in the central limit theorem then we have two hundred and something so we're probably okay now this linearity picture is not okay I was not linear at all I was like the best example of a nonlinear graph that I could have ever given you that's starting to be log linear so linearity would not be good so in general this is where I'd stop and maybe try to do to see if it's the zero one variable because sometimes that'll give you problems when you're using categorical predictors and then I might consider trying to do non linear regressions some nonparametric regression or maybe a log linear analysis okay are legit probit something different so normality is probably okay linearity is not blast to our homogeneity and homeless canasta city so for homogeneity what you want to do is look at the zero line you want to equal spread above and below zero and it's not equal because there's it goes up to four over here and down to negative two over here and they're clearly more dots over here and then the zero the other way that two to six just the same problem with our um normality graph so probably not homogenic and then homoscedasticity is that you want spread the dots to be even all the way across the graph and to always tell students to draw a line around the graph and it is not at all it kind of makes a fish and so that's not good for you either now there always be one or two little dudes out there so we always joke about those in class that these are delinquent children but well this is still even if I look at this this area it's not very good especially because of the cluster of dots down here so not the greatest example I would definitely probably try to do this different way more likely I tell you dude our use a different distribution family than the normal distribution like maybe pass on distribution or log-linear distribution but since this is a demo of process we're gonna keep going so there's my warning that I would not keep going on this but to show you how process works for three-way interaction I will alright so to actually run the analysis let's say everything actually does look good what you do is hit analyze regression and then process I think I have the newest version of process and this is SPSS 22 all right and so what I want to do is set this up so it looks a lot like a regression window just get more boxes and what we're going to need to do is use X is going to be our a CT score so our independent variable success is going to be our dependent variable so Y let's see here the GPA is going to be our M variable so our moderator and our second moderator is going to be W which will be ours intended if you stick it in covariates it will only include it as a main effect and while it looks like you can put extra things in in here it doesn't like it when you do that and so we're going to use M and W now X m and W are going to all be entered as main effects and all of their interactions so the order actually doesn't really matter that much but my idea theoretically is that the criteria for our scores should also dict the success in the course so this is the thing we actually use to determine whether or not students should go into that course and then it has some other variables I want to see if they moderate that relationship so the reason I'm in this order was more theory based all right so let's pick a model number we're gonna pick model three we talked about a minute ago so I'm going to do double moderation or moderated moderation or three-way interactions whatever you want to call it under options we're going to pick the first floor so mean Center for products to help us with multicollinearity problems I pretty much always leave on heteroscedastic consistence standard arrow errors especially because our data is heteroskedastic ordinary least squares on maximum life of a confidence intervals are always a good thing and then generate data for plotting which you can only use with the moderation models so we'll use that to make our picture of the data all right continue under conditioning I always like to ask for the Johnson name and I think it's really cool it only works for miles one and three and then we wanna you can change this pick a point thing the traditional way to do this for simple slopes is plus and minus one standard deviation but percentiles works as well you just have to tell people what you're doing so hit continue and then okay all right mmm so let's take a picture some of this and blow it up and talk about what everything is so the first thing you're going to get is just a reminder of what you entered let's go to this window and this a little easier to see because I can make it bigger okay so it's reminding me what I've stuck in each box basically and down to 227 after deleting missing people and outliers the very next thing down is the model summary box this is akin to the model summary box in when you run a regular regression and what that tells you is how much variance is accounted for by all of your variables including the interactions so we've got 11% of the variance so I'm going to look at R squared and then is that significant so my F value is significant so 11 percent of variance is greater than chance or greater than the mean of Y as the predictor and I would list the when riding it up as F and so it's seven and to nineteen which I got from that very first row DF 1 DF 2 is 4.6 5 P is less than point zero one and then I would do R squared equals point 1 1 and not in a talk so that's APA style the next thing that happens is we're going to get the B values so this coefficient here is non-standardized so that's B so we're going to get our regression coefficients for all of our variables and so my GPA variable here is not significant so it's not a significant predictor of their success in the course so that is still predicting line there a CT score is a significant predictor of success in the course so that is that for every for this is a categorical variable so for the one group they're more likely to be successful in the course their course scores one point higher than the 0 group not too surprising and my first interaction so interaction one is not significant and so let's see interaction one is a CT by GPA so there's no interaction between a CT GPA the hours that they study is significant but it's negative which is a little odd because people who are studying more should do better but it may be that the people who are studying more at the end haven't been studying the whole semester so I would need to look at like maybe some informal archival data to explain that relationship and then none of my other interactions are significant so I've got interaction to here listed here so it's a CT by hours interaction 3 it's really there's nothing there it's a GPA by hours interaction 4 is all three at once so a CT by GPA by hours wait so now those are significant which is why so this is not the best example but we could talk about each one of these at a time so I'll give an example let's say we want to list the coefficients for each one how would I write that up well the AC t-score you list because it's the unstandardized coefficient 0.0101 I list my T value the degrees of freedom for T always match the F statistic so it's 219 equals when I go over here to T so 2.09 then my p-value is 0.04 you can also us the upper and lower confidence intervals for B if you wanted so that's the lower level confidence interval and upper level confidence interval and it doesn't cross zero and that's why it's significant okay so we don't have any significant interactions but if we did so a lot of butts in this but you get the idea we would look at the next piece down which is the conditional effects so that's where it'll explain the interaction if there is one okay so what's going on in this graph it's actually going to give you um several different outputs but mostly you're probably going to want to look at the conditional effect of X on Y given the other two moderators so the very top part here is understanding that three-way interaction so it's got it broken down for you twice okay if you just want to look at the interaction between X and M only that is going to be this bottom part here and so you've kind of got the three-way part up here and then just that to the conditional effect of X and M given different levels of W so as a reminder X in our sample is our AC T cutoff score so it's a 0 to 1 M is our GPA scaled W is the hours a studied all right so how would I interpret this well what's happening is in this first box it's top box here it's going to tell me um it's going to create it doesn't really create group so don't think about these just groups but it creates slopes as if people were in groups for different levels so if I make a little table of what's going on this chart so there's going to be nine rows because I like tables if you've watched my videos for any of my videos for a while you'll see that I really like really just love tables all right so what's the first thing this is the gonna be hours here GPA and what happened so for low hours because that's the standard deviation is four point six seven so that's where it got a four point nine seven so that's what this number is it's of one standard deviation below the mean because these are mean centered so the zeros here are the average group and then the very next one over tells me I'm doing the low GPA so for a low scale GPA so two point nine nine points below the mean so if you wanted to know what the actual GPA for that was you would go back and calculate what's the mean of the score okay this is two point nine points below that mean so this is also low so for people in the low hours in the low GPA group what's the relationship between x and y so for our AC t group what's happening is that the effect is one point one two okay and that's not significant it's sort of marginally significant so AC t-- marginally predicts success and remember that in this particular example that means the difference between the two groups success scores so our our one group the people who have the higher AC t-- scores are doing better in marginally predicts success and all of these are positive so they're all scaled the same way so our groups are always doing better so the next column down is still low here because it's still the negative our score and then I look at GPA and that's average so what we're going to do is work through each combination so it's going to be nine of them so for a low hours group at an average GPA group what happens while a CT does predict predicts success and so the difference in the success scores is 1.30 okay so we're getting more more differences between our groups the next one I get low hours and a high GPA and our a CT score still predicts success and so the scores are going up to one point four nine this time so what does that tell me looking at this for people who are not coming to sessions as GPA increases the relationship between a CT and success increases and so as their GPA goes up and they're a CT score crosses this threshold the more like we are to be successful and then this is only for the low hours group though so what happens when we move up to the people who are coming to an average number of hours it's just going to be three of those as well so average hours and low GPA what happens so average hours low GPA the effects actually much smaller and it's not significant at all so there's no difference between the a CT groups so essentially as they come to more classes that effect the difference between them gets wiped out that's actually a good thing is they go to more of the study sessions they're performing at the same level as their a CT higher level a CT peers so at an average average group so that average average it is significant so there is now difference so while that helps the low GPA group from a low average and once we get into the average GPA group an average number of hours still we see a difference so it does predict success and our students in the above the criterion on the AC T are doing better so an average high now this time so average number of hours a high GPA average high it does predict success and the score is higher typing and so students are coming to an average number of hours and are in the high GPA group are doing better that and then there are HCT scores above the criterion are doing better but look at this overall the pattern we get to the same exact pattern that as they go as their GPA increases within the average group only for ours as a GPA increases the difference between my low and high a CT groups goes up which is the same effect here so these are increasing as GPA increases and so that's one of the reasons why we didn't get the interaction is because the pattern is the same while these numbers are physically different the pattern of the slopes is the same as their increase as the GPA is going up within each hours group we're going to get increase in in the relationship between a CT and success all right let's see the last group so people who are studying the most what happens so we're going to get high and then a low so what happens so people are going to a lot of sessions on below GPA there's no relationship between them so it does not predict success and then for a high group and the average GPA so high studying average GPA that p-value is not significant it's actually going down but it's still not there's no relationship and then for the high group of average I'm sorry the high group of hours and the high GPA so hi hi we get a marginal relationship but again look how small the difference is okay so it's um this actually is decreasing in slope so we didn't get the interaction with asked almost there because you're getting a difference the the slopes are going the other way for this group but it's not strong enough to show a significant effect and I feel like I mean I love tables but I feel like making this sort of table really helps you sit down and think about what's happening because three-way interactions are tough they're confusing and so this makes you break it down into groups and think about like what happens so for low average what is that GPA stuff doing so as GPA is increasing the relationship between a cc a CT and success is also increasing and that same pattern happens for an average group now for the high group there's almost no effect of a CT on on success maybe one small marginal piece but really there's nothing going on so if they're studying a lot their GPA doesn't matter if they're in the low average they're not studying very much in the low average group the GPAs matters a lot because the difference gets larger and larger as GPA increases and if they're in the average number of hours studied group GPA matters but the difference between the low and a high is not as strong and so thinking about like what's happening at each of these levels the only thing you can interpret here is this conditional effect of of X and M interaction at different levels of W so this sort of collapses across GPA and looks at only hours so I found this a little harder to interpret but at a low number of average hours the overall effect of GPA is point oh six mm okay I'm sorry that's not just GPA at a low number of hour at a low number of hours can't talk today the interaction affected GPA and a CT is 0.06 and that's not significant so this is this number here isn't one particular variable at a time it's interaction of XM together add an average number of hours X and M together are still not significant but it's point o2 and so this is actually changing directions and so on a high number of hours it's actually a negative interaction this is nice to have but it's this to me is easy to interpret the conditional effects of XM why because I can deal with one variable at a time if ours is low and in GPA is low what happens between the relationship of a CT and success so I've been a little easier because then I can I can collapse if I want to so what happens at only low hours but mmm interpreting these effect B effects for interactions can be kind of difficult okay so that is how I would interpret the output now we didn't get a Johnson name in because of the non significance here but what the Johnson Eamonn will do is take this conditional effect box that we've spent a lot of time looking at and break it down even further so you don't just get low average of high you get different points along the continuum so you get more of these significant points and it will tell you where things are an art significant so it's really fantastic if you watch my my single moderation video I have an example of the Johnson name it in there okay so the last piece is this graph now the newer versions of prot stupid morning the double click to activate makes me crazy all right where we go all right so the newer versions of process which I apparently don't have installed on this particular computer actually will give you some and syntax to create these graphs but creates you scatter graphs and so I always do these by hand instead because it tends to create a picture that I was more interested in and so we're going to do is take this data and enter it into SPSS so I'm going to create a new SPSS window here one new output a new window new data set pile it up next to word so what I'm going to do is instead of using these numbers I'm just going to call it low and high so this first column needs to be let's go a CT score labeled here oops okay and then we're gonna do our our score so this is hours of study sessions and then our GPA which is scaled GPA since schools use different systems and then our Y variable which is success course success good thing I'm not known for my spelling all right now what I'm going to do is create low and high values and since a CT is only a zero on one score it's only given me it's only given me zeros and ones for that if it were a real continuous variable it would give you low medium and high for or low average and high at that level as well so in this particular example I'm going to 0 for low or actually in this for my example I could to do below criteria because that's specifically what we're doing but you could do low and then above criterion what I'll do is enter one zeros till we have enough columns that we need even turn on value labels here so you can see that a little better and that's what's this is representing this is the low group to high guerilla group the high group we're going to need 18 of those and I find that easier because then it's labeled for me on the graph this one reason why I do this um you can use these raw numbers but they need to go back and change them to be labels so under ours here I'm going to do negative 1 for my low group 0 for my average group and one for my group I like that coding because that's what those ours is one standard deviation above and below 0 but that also means that I can cut and paste it into my GPA column as well so I put that in ours and GPA it's what I want to do is go back in there so I've got lo looks like six times oops I did I 6 okay so that matches this low average high now in the GPA column it's low low average average high high so I don't want to make sure I don't enter the exact same column twice for ours and GPA or your graph will not work and you won't know what's going on so they will have different patterns in each column and so these are the we're just breaking it into two groups in a sense to be able to create this graph picture in the data is really continuous but this will help us make a picture a graph and then I would want to enter the Y hat scores so these are predicted scores for people on each one of those groups and this is where you got to type the numbers so we'll do two decimals just to make it go kind of quick here's bad typing things on video if I am definitely get people to check your work our graduate lab has a whole wall of funny things that we have written in reports and not realize because we're all we all make mistakes okay I got a line off somewhere so I got four point three four or five four three point nine two that is so I got three a five and I skipped one here it's four point eight six let make sure I hit the rest of them right perfect okay so now how do I create this picture we're knees tribe builders so graphs chart builder it's really grumpy with me because I haven't set the variable property so I go back over here under a CT score under measure I'm going to make that one's nominal and then make the rest of them scale all right now graphs under chart blur yeah yeah all right so here's chart builder I'm going to make sure that you have these little variables type of metric variable on there will force you to do that first we're going to pick a line graph and I'm going to use a multiple line graph and it's actually not going to let me create a three-way interaction in one graph so what we're actually have to do is back out of this first and do one extra thing but let me show you let's go ahead and kind of set it up and you'll see why you can't do three through interactions in one graph so I could take I won't always usually put x and y in there places that I thought they were so success was fine and X was my AC T score but I might split my data on X so let's actually put a GPA in X and hours of study in setcolor which means I'm going to have to force that variable to be a a nominal variable so you can right click on them i'ma click nominal just to make sure it I can go in to set color because set color requires it to be a nominal variable now you can move these around the Quartus sort of create the best picture generally I suggest to put X here and Y here because that way it matches the simple slope so that you are doing but in this particular example because X is a a dichotomous variable you might consider splitting everything on X and that'll make more sense here in a second but if I wanted to do it the way I've told you to I've told everybody else to do it I might put my X variable here and then actually my M variable my scaled GPA here a lots of different ways to do these graphs so it's kind of to you so let's hit OK and that will give us the graph for above and below criteria the relationship between GPA and success in the course but now I'm sort of stuck because I don't have hours so what you want to do is go to data and then split a split file and let's split on hours so organized output by groups and then move hours of study into here so you could pick any one I could split on a CT and look at people who are just below my criteria and people are just above my criteria I could split on GPA so let's look at low average and high GPA or just split on ours so it's kind of depends on how you want to sell the story and how you're going to talk about it in your output so whenever you're writing this up split based on the way you discussed it in your results section so it's a mate okay and now let's make that same graph so graphs chart builder okay and then I want to leave this actually the same so don't split on a variable and try to use it it'll freak out but I'll leave all this the same and hit okay I'm going at three different graphs I'm gonna graph for low average and high generally it tells you oh here this is it have study sessions low so here is the relationship between GPA and success in the course given a CT score for low people only and so you can see that as GPA is GPA increases it seems to matter for the people in the below the criteria but above the criteria not so much and then for an average number of group would get almost perfectly parallel lines so GPA is making a difference whereas the the low people are actually doing better in the course which is sort of weird and so maybe but that study sessions are working like we expect them to remember this is not the best dataset and people who are above criteria doing better than below but there's no interaction because the slopes are parallel and we see it almost the same picture here so this is the one that actually has slightly that the slopes were getting closer to zero and so they're a little flatter so you're going to end up with at least three graphs maybe more if you have more than if you decide to do like two standard deviations above and below the mean or if you're doing some Johnson name and graphing but that's probably the easiest way to graph them given the output that you get for from process and so that worked through the whole thing on a three-way interaction or moderated moderation using model three in the process plug in from data screening all the way through graphs and then let me know what questions you have

Channel: Statistics of DOOM

Views: 42,805

Rating: 4.9352226 out of 5

Keywords: SPSS (Software), Statistics (Field Of Study), moderation, regression, three way interaction, interactions, outliers, data screening, mahalanobis, cooks, leverage, normality, linearity, homogeneity, homoscedasticity, process

Id: 4k8JFvJRw_w

Length: 53min 5sec (3185 seconds)

Published: Wed Jul 08 2015