morning hello I thought we could do with some Slayer to get us in the mood today because you know it's week as job loss Cal more week is maybe 10 possibly and you know the the forums are getting a busier people are seeming more on the edge so that's normally when Slayer is what you need to but maybe take you over the edge but mainly pull you back who knows um but also the tenuous link to the lecture is that was a song about a serial killer called Ed Gein and he's going to pop up a bit later on on the screen that kind of rhymed so today we're going to carry on the an OVA theme um last week we did an Cova did we how if I'm losing track and and this week we're moving on to well we do it's kind of variations on a theme so at some point I did say you know from now for the rest of time it's it's kind of all quite a similar territory and I'm going to attempt to be true to that promise by just sort of doing more an over really so it should all work out good hopefully you never know so what we're going to do today is have a look at what's known as a two-way anova and what I explain what that is in a minute but really it's it's just it's quite similar to the an overs we've done before we're just kind of moving on to look at when we have more than one independent variable or if you want to put it another way more than one categorical predictor so will it very very very very briefly look at how this affects how we partition variants but we blink and you'll miss that and we're not going to do too much theory at all we're mainly going to concentrate on the pragmatics of how this differs from other an overs look at two examples how you would interpret them and sort of things so there's going to be a lot on interactions now we've already come across interactions when we did the lecture on moderation which was in week something else and it was just before your class test so I think that was the lecture and it was about two people here so those two people and think back to moderation because we're continuing that theme so what is a two-way independent an / well when you read journal articles your you'll see that an / is used a lot because in many many situations psychologists will be manipulating independent variables using different groups of people or as we'll come onto next week subjecting the same people to different experimental conditions so dealing with groups dates are also categorical predictors is pretty common so when you read research articles you'll quite often come across them saying we did an ad over of some description now the type of an odor that people do depends on the design of their study so so far we've looked at where you only have one grouping variable so you've manipulated one independent variables and you know we've dealt particularly with the situation where you put different people into different levels of of that independent variable so in our viagra example we had a libido group a low dose group a high dose group containing different people so that will be known as a one-way ANOVA and the one-way just refers to the fact that there's one independent variable so when you come across someone having done a two-way and over what that means is they've manipulated two independent variables and you might also come across studies with of they've used 3-way an overs and that's where they've manipulated three independent variables so whenever you see an ANOVA described or if you're trying to describe the an over that you've done on your data you need to use this or one way to a three-way description and that just refers to the number of independent variables you've manipulated now as you can imagine if you start manipulating more or independent variables that the the analysis gets quite complicated so on this module we only ever look at a three way ANOVA that's the last letter like a Christmas present to you all is to look at a three way ANOVA in the very last lecture and that's as far as we go and the reason why we only go that far is because I think if you have more independent variables in three your analysis starts to become completely uninterpretable which is not to say that people don't do it and in fact I have done I've got papers where I've done four and five way and others but in retrospect they were stupid things to do I shouldn't have done it so there's a sort of there's a pragmatic reason in that you know I think a three-way three-way Andover is about as complicated as you ought to be getting as a scientist and if you're getting more complicated than that it's going to be very difficult to actually work out what's going on the other thing is your notice in descriptions of a Nova's people often refer to them as being say independent an overs sometimes people use the phrase between groups or between subjects as well but basically they're all the same thing as an independent ANOVA so independent just means different participants were put into like different groups so the variables the independent variables were manipulated using different people to represent the different levels of those variables so when you see an independent ANOVA or a between group and over or sometimes even a between subjects ANOVA they all mean the same thing it just means different people in different conditions the other term your some type will come on to this next week you'll sometimes hear Andover's referred to as repeated measures ANOVA x' or within subject and overs and that's where you've used the same participants but tested them under different conditions but that's that's for next week so at the moment we're just going to deal with a two way ANOVA and we're going to deal with an independent design so this is where all of the various conditions that we have we've used different participants in to those conditions now where we've got more than one independent variable you'll sometimes hear it referred to as a factorial design that's why the lecture is called factorial ANOVA so that just means you have more than one independent variable so when you I mean this is all basically like years and years ago someone decided like some BD statistician thought you know you know what I've just I've just invented the central limit theorem or something like that I think statistics is getting a bit too easy for everyone to understand what can I do to confuse them I know I'll invent lots and lots of different ways of expressing exactly the same thing so you know rather than calling something you know an independent design let's make it acceptable to call it between group and between subject as well so there's three different terms that mean exactly the same thing and you see this or thing time and time again in statistics like that well I've touched on it before actually with error and residual and things like that you see the sort of different words referring to the same thing a factorial design is is one of those kind of things because sometimes independent variables bizarrely are referred to as factors so that's why it's known as a factorial design but this gets very confusing because next term you're going to learn about something called factor analysis and the factors in factor analysis are completely different to the factors that we're going to talk about now which are independent variables so it's it's it's a linguistic mess for us all but you know that the touching a nice point to take from this is that it's confusing for all of us not not just you so why would we do a factorial design why would we make things more complicated that's it you know that's the same thing to do showing you want to keep things simple so that's is hard enough as it is why make life more difficult well the main reason is because we can look at how our predictor variables our independent variables interact with each other and this is a really really powerful tool and this goes back to this idea of moderation that we dealt with earlier on the course so interactions between variables show up moderation effects what a moderation effects that is where one independent variable has an impact on the effect of the other independent variable so these are basically these moderation effects are more often than not more interesting than looking at the effect of a variable in isolation so what do I mean by a moderation effect well one very simple example I teach two very different courses I teach this course module sorry and I teach a third year option on anxiety in children which is why I do research on very very different topics as you might imagine and you might also imagine that from my point of view very very different student kind of feedback because I'm off fear and anxiety course you know I could literally just sort of turf up and ramble on in a fairly bored way and everyone was oh it's a really interesting course because talking about how anxiety develops and children is intrinsically a bit more interesting than heteroscedasticity apparently whereas on this course module whatever they're called these days you know people come into it with the expectation that it's going to be very dull so you know I feel like I need to make more of an effort and so students have sort of different reactions to the courses they still come into them already you know expecting one to be boring and expecting the other one to be interesting but the other thing is students also you know looking out 9 o'clock on a Monday I actually have my third option immediately after this I've got a very direct comparison of what a group of students at nine o'clock and a stacked lecture look like compared to a group of students in a in a clinical psychology lecture at 10 o'clock on the same morning and typically the ten o'clock ones they still look about as bleary-eyed as you do to be perfectly honest but they they do look a bit less scared as a general rule and but anyway I might want to look at say you know if you've been out drinking the night before that's going to affect how sleepy you are in a lecture so I might be interested in that effect so is you know because I'm kind of a paranoid person if people are sleeping during my lectures I tell you know I think I'm being dull and I don't like the thought of being dull which is slightly odd given why I teach but anyway and so I might want to test that is it the case that people sleep more when they're hungover than when they're not so that's the effect I might be interested in but I might want to you know take account of the type of thing that I'm teaching so it might be the case for example that because stats is a naturally less interesting topic that also has an effect on whether people sleep or not maybe people are more likely to be awake if I'm talking about how anxiety develops then if I'm talking about moderation effects so I might want to combine those into a single study and it might be the case that for example the sleepiness of you know how sleepy you are is not particularly affected by how hung over you are unless you happen to be in a stats lecture so that would be a moderation effect so having a hangover doesn't affect your sleepiness in a clinical lecture so in a clinical lecture you sleep about the same or in general sleepiness levels are about the same but in a stats lecture there's a suddenly a big difference like if you have a hangover that's it you know the mention of skewness just you're off like narcoleptic fit and so that will be an example the moderation effect now I'm going to label this point and go through a different example so the other thing I stumbled across a I'm always stumbling across papers that are not remotely related to what I do research on like somewhere across this is paper about injuries from playing computer games so it's a little montage that I made of some console related injuries this is my favorite can you see where this one's going so I'm yes playing computer games may well injure you or your nearest and dearest so we could put this to the test we could look at one independent variable which was the type of console so are you playing a Wii or are you playing an Xbox because I they have different ways of doing the motion by I'm a bit out of my depth here I think really do games consoles but now different ways of doing the the motion thing so from what I gather on a wii you have a little remote control thing that you can bash your children around the head with whereas on an Xbox it does some magic where you just go like that and it detects your movement um so you could try different games consoles but you could also try yo static games versus you know active games there are some games you know where you're just sitting there shooting stuff or shooting stuff she's stuff I don't know what you do but there'll be other games you know like the I assume he was skiing the guy who trod on his dog when he was doing all that something I don't really know and so you could have a look at active games so you got two independent variables here one is Xbox vs Wii and the other one is a static game where you're not moving around doing stuff this is one of those kind of active games where you're you're using the movement technology to play tennis or or just beat up your children so we could have a look at what known as main effects so this is the effect of an independent variable in isolation so for example this would be the main effect of playing a Wii versus playing an Xbox so we've got the severity of injuries that's our outcome variable and we could have a look at well does it make a difference whether you whether you're playing Wii or Xbox and actually from there's a bit of a difference here but it's not much of a difference so there's you know kind of probably not too much of a difference in injuries based on the console you use what about the type of game so a static game versus an active game so here we've got a bit more of a difference and the error bars are not overlapping so that's probably significant so we could say well there's a main effect overall the injuries are worse they're more severe in the active games that's the dark blue bar over here then in the static games as you might expect when people are moving around they're more likely to to get injuries but what we're interested in is again this interaction or the moderation effect so are there kind of differences in the effect that static versus active games have depending on which console you use and that would be shown up when we look at all of the conditions together so you know recall here we've got the static games and the light bar is the Wii and the W on it draw and the dark bar is Xbox so in the static condition so when they're playing static games games where they're not moving around using their kinetic technology or whatever it's called there's basically no difference between the two consoles it doesn't matter whether you're playing a Wii or an Xbox the severity of the injuries that you sustained as a result of playing those games are pretty similar however in the active games so when you're actually using these games where you're supposed to you know dance or ski or all or throw your telly around there suddenly does become a difference between the two consoles so again we've got the Wii over here and Xbox over here and you get more injuries from the Wii so when you have a you know a thing in your hand that you can throw across the room you get you get more injuries or more severe injuries than you do with the Xbox so notice that in the active condition both of those bars are higher than in the static condition so for both the consoles you get a bit of an increase in the severity of injuries for the active condition but that's particularly pronounced for the Wii so for the Wii the bar is higher than the Xbox so this is a classic moderation effect in the static condition no different I mean I've exaggerated it a bit there's literally no difference between wheeze and Xboxes but in the active condition you suddenly get a difference between the two consoles so the effect of the console is moderated by the type of game that you're playing or you could you could actually you could put it the other way around as well so the the effect of the type of game you're playing is moderated by the console so that's what we're talking about with moderation and that's really what this lecture is all about so we're just going to go through a couple more examples to sort of drum the point home so to keep things close to the textbook and the handout the first example that we'll go through is about the beer goggles effect which we talked about a bit in the lecture on writing up lab reports so this is the idea that your well your sort of standards for own choosing dates are affected by alcohol so the more you drink that you get the beer goggles on and you start losing all sense of what's attractive and what's not so we've got an independent variable here that represents the beer goggles effect which is a dose of alcohol so if you've had no alcohol versus say two pints of regular strength something or other vodka or four pints of regular strength something rather and if your outcome was you know the attractiveness of the partner that you chose on that night then that that effect of alcohol is basically going to tell you the beer goggles effect now what we might want to do is do to say well is the beer goggles effect going to be different in men and women that's again a moderation effect so are males more or less susceptible to the beer goggles effects than females so to test this we'd have to also measure the gender of the participants so we've got two independent variables a dose of alcohol and the gender as I said our outcome variable would be some color you know let's say we sort of dosed them up so we've got different groups we give them different dose of alcohol obviously the males and females will be different groups you're either a male or female so that's self selecting and your outcome at the end of the night they all go off pick a date and then in a slightly hideous way with as they go out we have a panel of judges to judge the attractiveness of the maker they selected very ethical so the data might look something like this now it is an important service or inwardly digest these data particularly but just note that we've got six different groups representing all the combinations of our independent variables so these females had theirs or two forces there's eight females here who had no alcohol there's a different eight females here who had two pints because it's an independent design and a different group of females here who had four pints and also we then got three groups of males so a group of males who had no pints a group who had two pints and a different group again you had four pints and these are the attractiveness ratings of the partners that they chose so that's basically what we have six different groups of people for each person in the in the study we have one rating which is a rating the attractiveness of the mate that they've selected so what happens when we partition variants now I said I go this briefly and I will so when we do like a one-way on over or when we fit a you know a regret a linear model of any description we've seen before that will have some kind of total variance so that will be in this case the total variability in the attractiveness ratings and we slice it into two so one of those slices represents the variance in those ratings that's explained or accounted for by our experimental manipulation now in this case we've got more than one experiment to manipulation so the only difference in this diagram to ones before is I've put an s on the end there those manipulations not manipulation the same principle basically we've got a chunk of variance that's explained by the things that we've done to the different groups and we've got another chunk of variation over here which is the error so that's just variation in attractiveness ratings that can't be accounted for by the variables that we've manipulated so up to now this is exactly the same as what we've seen before for any linear model that we've covered that's the basic process you got variance accountable by the model versus error the only difference here is that this variance accounted for by the model gets partitioned up into an effect of your first independent variable in this case alcohol and affected the second independent variable in this case gender and the effect of their combined effects if you like to say their interaction so we just when we have sort of more than one independent variable we just get out our sort of experimental variance born of a better word get sub divided up into you know from different effects so effective one independent variable the other and the interaction so it's really it's no different to what we've been doing before the only difference is is rather than getting one effect so when we've done an overs before whether it was in the context of regression or analyzing different groups we've always had just one F ratio but now we're going to get three we're going to get an F ratio for the two effects of the independent variables and one F ratio also for the interaction so that's the only difference to what we've done before is just you can think about it like we have every other linear model we're just adding in new predictors so when we when we looked at moderation we talked about adding in sort of adding in new predictors like in a multiple regression we're just doing the same here rather than having one categorical predictor we've got two categorical predictors and their interaction as a predictor as well so it's all it's just that is exactly the same as what we've done before we just get more FS in our table so when you do this in SPSS you get a table like this which looks a fair bit more hideous than some of the tables that we've come across before but it becomes less hideous if you do this and basically ignore the stuff that you don't actually need to pay any attention to so there's three things in here really that we need to pay attention to your three rows of the table one of them shows our what's known as a main effect so it's the effect of gender on its own the main effect of alcohol that's the effect of alcohol on its own and also the effect of the interaction so everything else in the table we can pretty much ignore this is the same as every other ANOVA table that you've come across on this module so we've got an F ratio for each effect and that F ratio is calculated by taking the mean square for the effect and dividing by the mean square error for the model so the EPS compute in exactly the same way it represents exactly the same thing it's the ratio of the effect to the error in the model and that's true for all three of these so you get an F for each one of them and you get a p-value for each one of them as well so we're just looking where those p-values are less than 0.05 so in the case of gender it's not less than 0.05 so that would be a non significant effect but for alcohol and for the interaction those p-values are less than 0.05 if they're quite a lot less they're very significant so basically we've got no significant effect agenda we've got a significant effect of alcohol and a significant interaction all of these effects have degrees of freedom associated with them so if you're reporting the effect of gender you'd want the one degree of freedom and the aro degree of freedom if you're reporting alcohol you'd have the degree of freedom for alcohol and again the aro degree of freedom and if you're reporting the interaction the degree of freedom for the interaction and again the error so with all of them you report the error degrees of freedom and the degree of freedom for the effect so it's all very straightforward very similar to what we've done before nothing particularly new or or worrying or anything so let's have a look at what these effects actually mean so what does it mean to say there is a main effect of alcohol so we got a significant effect of alcohol what does that actually mean well imagine that we had not measured gender at all so we just forget that information ever existed so all we know is that we have a bunch of people who had no alcohol a bunch who had two pints and a bunch you had four pints so notice I've put them in two boxes and the scores in the boxes represent all the scores of anyone who had no alcohol so this first column is actually men and or they might go around a can't remember and the other column is women but we don't know what they so we just imagine we don't know that and we've just lumped them together in a group so we've got all the people together around no alcohol irrespective of their gender and put in a box put their scores in a box then we've done the same for people who had two pints and the same people who had four pints and then we calculate the mean of those groups so people who have no pints irrespective of their gender the mean was six world just under 64 when there are two pints is just over 64 and after four pints it was about 46 and a half so the main effect of alcohol is like doing a one-way anova on those three means it's exactly the same you ignore gender and you just what effectively what's going on computationally is it's doing a one-way ANOVA to test whether those three means are different so what you might find is well we did find actually that they are different so we can report that as being there's a significant main effect of the amount of alcohol consumed on the nightclub at the nightclub on the attractiveness of the mate selected we report our F our degrees of freedom as I just told you what they would be and the p-value which in this case because it's so small we can report it as less than point zero zero one what that tells us is these three means are somewhere along the line not equal to each other so that significant effect means that you know amongst those three means there are some differences we have to go and do some post-op tests or something to work out where those differences are we just know they're different so looking at a graph it becomes reasonably obvious where the differences are so we can see that the means for the no and two pine groups are pretty similar there's not much of a change there at all but there's a sudden sort of dramatic decrease in the means for four pints so although we need to do some formal tests to demonstrate this basically this main effect of alcohol is saying that when you ignore gender the beer goggles effect kicks in at four pints it doesn't kick in a two pints because two and zero pints is all the same it kicks in at four pints so four pints suddenly you're the attractiveness of the person you select dips what about the main effects of gender will effectively would do exactly the same thing so we're pretending that we don't know how much alcohol people drunk and we're just grouping them according to whether they're males or females now this should be the point at which you start to realize why the interactions are more interesting than the main effect because at this point when we're looking at the effect of gender when we're looking at whether males differ from females we've got some females here who had nothing to drink we've got some who had two pints to drink and we've got some you've got four pints to drink so that group of females is is not I've sort of a very homogeneous group some of them are shit-faced and some of them are completely sober and the same is true in the men some of them these guys are sober these guys have had four pints so although we can say you know there's a main effect of gender whatever you've got to bear in mind that when we've lumped all the females together those females are you know have different levels of intoxication and the same with the males that male group is made up of you know differently intoxicated men and although they'll sort of balance out in the sense that you know you've got as many women who had four pints as you have men it's still you know it's not really comparing gender on it on a very even keel because you know there's a lot going on within that group but nevertheless if you wanted to interpret this main effect basically the F ratio that you get is going to be testing whether those two means are different and we saw that this wasn't a significant effect so our p-value was equal to 0.16 one that was non significant so what that means is that a mean of 60 point to one is equal approximately to a mean of 56 points for six in other words on average the rate they are attractiveness of the mate selected by females was no not really different to that of males again you can see this very clearly on a graph so the F is just telling us are those two means different you can see they're really really really similarly arrow arrow bars completely overlap so nothing much going on there now what about our interaction this is the main thing that we're interested in so how do we go about interpreting these well essentially you can do it in one of two ways the first way is to say what's the effect of gender at each level of alcohol so we could here say well when they've had nothing to drink what's the difference between men and women so there I've been a bit gender stereotypes here in giving the male was a blue line and the females us or pinky red line and so what's going on in there after after no alcohol well basically the ratings of the male partners are slightly higher than those of the females but not very much they're basically sort of the same the error bars are overlapping so there's not really a gender effect at no at no point what about our two points is there a gender effect there well again we get basically a very similar result so the male and female or the attractiveness of the mates that males and females select are basically kind of the same they're more or less equivalent so at no points and at two points there sort of isn't a gender effect now what happens at four points what are four points suddenly we start to get an effect of gender because the mayor though the rate the partners that the males pick their attractiveness is much lower so the blue line is much lower than the red line so at four points you get this very sort pronounced effect in males there's a big dip in their in their quality control but the females there's not really an equivalent dip so you can this interaction whether the significant interaction represents the fact that they're kind of isn't a difference between men and women when they're sober there isn't really a difference when they've had two points but when there's four points there is a big difference so it's like the beer goggles effect kicks in for men at four points but doesn't for women now the other way you can look at that is to look at the effect so we've looked at the effect of gender gender at each level of alcohol you can do the reverse you can look at the effect of alcohol at each level of gender so we could say well what's the effect of alcohol in the females so what's the pattern in the red line or the pattern in the red line is it's pretty flat it doesn't really change too much which suggests that the effect of alcohol in women is is kind of no effect really they're there the attractiveness the partners they pick is relatively unaffected by alcohol what's the effect of alcohol on men well again we get a very different pattern so at low doses of alcohol nothing much is going on but you suddenly get this dip at four points so the effect of alcohol in men is different to the effect of alcohol in women so it basically it doesn't affect women does affect men so that's what an interaction is and that's how you need to try to solve it interpret these graphs so we'd say there's a significant interaction between amount of alcohol consumed at the gender of the person selecting a mate on the attractiveness of the partner selected again report the F degrees of freedom and because the significance was was like zero we can report it as less than point zero zero one sometimes you'll see interactions expressed as bar charts rather than line charts but the same basic principle applies so here we've got a males as the light blue and females as a dark blue bars so again you can see not really a gender difference for no alcohol not really these are the same data not really a difference after two pints but at four points you suddenly start to get a difference between the genders I just wanted to put a bar chart up because you know some sometimes people do bar charts sometimes do line charts you wouldn't ever do both but just you all need to be able to interpret both because you know you can do either so you'll come across either when you read research papers so what is an interaction well just to simplify things let's imagine that let's imagine that we just had a scenario where there were no pints or four pints so we've got rid of the middle condition just to make things a bit easier so what an interaction is testing quite literally is so first of all we have a look at the difference between men and women after no pints and that difference turns out to be 6.25 so the male's of Exarchate of us what the colours around here which is not very sensible things too so the males who are now the red line i've got higher slightly higher attractiveness ratings than the women so that's a difference of six point two five after four pints if we look at the difference here now notice the lines of crossed over so now the male line is below the female line so that's actually a difference of minus twenty one point eight because they've switched over when we do this subtraction it ends up being a minus number so applied these two numbers here so this this number here is six point two five and that number there is minus twenty one point eight the interaction is literally testing is six point two five different from minus twenty one point eight so are these two two values down here different from each other and if they are then you know you'll get a significant interaction so it's literally testing is you know is the difference between males and females at no alcohol different to the difference between males and females at four points here's an example if we didn't have a significant interaction so let's imagine for no points we've got the same difference as before so six point two five but now notice the lines are parallel to each other so after four points nothing much has changed really so we've got difference between males and females of only five point six this would be a non significant interaction because the difference between men and women at no points is basically the same as the difference between men and women at four points now the important thing about this diagram is notice that so on this side we've got a significant interaction and what we find with a significant interaction is is our lines of gender lines of kind of crossed over so typically if lions cross over on a graph like this that might indicate that the interaction is likely to be significant won't always be but it's a it's a you know a reasonably good visual guide if however your lines are parallel to each other that basically means there is almost certainly not going to be a significant interaction because of the nature what the interaction is is it's testing whether the basically the gap between the lines here is the same as the gap between the lines there so the lines are parallel it means the gaps are not really changing okay let's have a look at a different example I don't know is this the music stops working normally when I do the penny thing but you know what it's probably good thing because this was a clip of Robbie Williams which you know I put in just because this is an example about Robbie Williams but if I'm honest I don't really want to hear it because it's week ten we're all emotionally vulnerable that's the last thing we need so this is an example about and because I am Wow it's not you know I don't know Robbie Williams this could be it could be a lovely guy I'm sure but I find him quite irritating if I'm honest and there's not how many people are find irritating in general I try not to be irritated by people but Robbie Williams does definitely push my buttons in certain kinds of ways so I'm in my fantasy world where basically everyone is a clone of me everyone would dislike where we live that's clearly not the case because you know he sells out stadiums to adoring fans so young suspect I may be in a minority but I wanted to test whether attitudes to Robbie Williams were you know negative or not so I came up with this study it's a priming study I just took took a paradigm from social psychology and what I was going to do is I was going to prime nice or nasty personality characteristics so some nice and nasty personality characteristics are going to pop up on a screen but I'm going to present them really rapidly so they're sort of subliminal so you have to try really hard to detect them so my question is what happens if you prime these nasty characteristics with different pictures so I decided to use Robbie Williams so the idea is that if you put up a picture of Robbie Williams if you like Robbie Williams and then a nice characteristic peers on the screen you should be better able to spot that characteristic because seeing Robbie has put you in a kind of nice you know it's activated you're nice person Skiba and therefore when you see the nice characteristic you can attend to it much more rapidly and actually spotted before it disappears so I had a group of people who's just saw pitch with Robbie Williams priming nice personality characteristics and they just had to recognize these very rapidly presenting words then I has a different condition we're still pictures of Robbie Williams but it's priming nasty personality characteristics so the idea is you know if you if you don't like Robbie Williams then that will activate your nasty person schemer and so you'll be better at detecting these lasting characteristics now I needed a control for Robbie Williams and I thought the best control would probably be Ed Gein who was an American serial killer and for if you've ever watched the Texas Chainsaw Massacre that's loose loosely based on the story of Ed Gein he was a strange chap it would be fair to say who was I mean you know there's lots of serial killers but his his particular thing was peeling the skin of his victims and making like you know dresses out of them and dancing Haley's kitchen and stuff nice nice go I was in a bar and so the thing is I'm worthless was a good control is I said Robbie Williams used to have a video I think it might be for rock DJ I'm not sure anyways poncing around and he saw he peels his own skin off and starts throwing it around so I thought well you know it's a good control for sore skin peeling if I use Ed Gein the serial killer also you know Ed Gein you would kind of university expects everyone to think he was nasty because he was a psycho anyway so everyone saw 30 trial was effectively so 30 target words that they had to try to recognize and there were seven people in each condition so what the variables I've got here when one of my variables is going to be the type of prime so whether people saw pictures of Robbie or pictures of Ed gay I'm going to have an independent variable which is the target so this is the the word that they had to recognize so for some people they were nice words for some people nice personality characteristics and for other people they were nasty personality characteristics and my outcome is simply how many of the 30 words could they actually detect because remember I'm presenting them really really rapidly so let's have an example of a trial so what they would see is a fixation cross to fixate their attention on the middle of the screen and then then see a picture so this is a picture of Ed Gein and then after so this would be so displayed on the screen for a relatively short period of time and then there'd be a rapidly presented word that would then be masked to make it difficult to see so watching did anyone see the word we're doing slow-motion so fixation cross picture Ed Gein then sort iron water true generous and then it gets masked so the question will be on this trial could the person say the word was generous and like I say the idea will be if you if Ed Gein is activated a nice person schema it will be relatively easy for you to detect generous than if that hadn't happened so let's have a got another trial here's another picture of Ed Gein slow-motion again so it's been nasty characteristic selfish so again the question is can you recognize his characteristic so let's try robbing trial so you fixated Robbie flashes up on the screen anyone spot the word no Robbie fans in here then so that would be slow-motion again egotistical with Robbie picture flashed up on the screen maybe another trial with Robbie anyway so you get data a bit like this so again four groups of people different people in each condition so some of them saw Robbie priming nice traits some of them saw Robbie priming nasty traits some sort ed priming nice straight some sort ed priming nasty traits and these are how many words out of 30 they managed to recognise so again we get a similar table to the one we had before so we've got three effects and effects a prime and effective target and the interaction so we're looking at I've lost me I've lost my ability to point those stuff on the screen I'll point to this so we've got p-value of 0.02 4 4 prime so that's less than 0.05 so significant for target we've got a significant effect because 0.01 3 is less than 0.05 and for the interaction that is also significant because point O 3 2 is less than 0.05 and here are some means what I'm going to put them on a graph in a minute so we've got a main effect of prime so this is when we ignore whether it is a nice or nasty personality characteristic how many words did people recognize and you can see from this that that I mean it was a significant effect because the p-value is less than 0.05 but we can see basically that Robbie up here the mean is higher than edge so people when we ignore which type of characteristics they were looking at people recognized more words when they were primed with Robbie Williams than when they were primed with Ed Gein but I can say in a way because there's a significant interaction we don't we're not really interested in interpreting this because we know that this this bar here is made up of both conditions the conditions that have nice characteristics and nasty so this is a sort of mishmash of stuff and so you know in its own right it's not that interesting the main effects of target was also significant so this is nice characteristics over here nasty characteristics over there so again this is telling us if we ignore whether Robbie or ed was used do people recognize more nasty characteristics than Knights and the answer is yes they recognize significantly more nasty characteristics than nice ones but again this isn't particularly interesting because we know that this mean here is made up of the conditions that had edge and the conditions that had Robbie so you know there's again a mishmash of stuff within this sort of bar so what we're really interested in is this interaction of Prime by target interaction does anyone want to have a go at interpreting this effect so what you need to think about is what well one thing you can think about is what's the effect of the different characteristics for each of the primes that'll be one way to tackle it I didn't bring any chocolate today it's an era yeah fantastic so for Robbie we get this big difference where they recognize more nasty characteristics than nice characteristics now what what's going on for eight anything different no difference so there's kind of no difference for Ed Gein between the number of nice and nasty characteristics recognized but there is a big difference from Robby's so when when Robby is being shown people are recognizing more nasty characteristics now you know the psychological implications this is that they hey Robby is the more hateful character than Ed Gein the skin peeling serial killer like I said data from a fantasy world that only I occupy probably and the other way you could look at this is to look at what's the effect of nice characteristics and the effect is the same for Ed Gein and Robbie Williams the lines flat and what's the effect of nasty characteristics well it changes depending on who's priming it so you get more recognized for Robbie than you do for Ed so you can look at it both ways you can either look at kind of the effect of characteristic at each level of young prime or you can look at the difference in the primes for each level of nasty characteristic this is just showing the same date with a bar chart again just because I want to get used to the idea that you might see these interactions as bars rather than lines and again this shows a very clear picture in that in all these conditions you get about always about 10 out of 30 of the words recognized so about a third of the words are recognized in every condition apart from one and that one condition is Robbie priming the nasty characteristics so on balance across lots of open pretty much all the conditions people recognize about one in three words apart from when it's Robbie priming a nasty characteristic they suddenly can recognize more about half in fact so just by a show of hands without any other information who reckons that that would be a significant interaction to be free for more confidence is growing ok quite a few of you that's very good that would because the lines are not parallel that would be I don't guarantee you but it's likely to be a significant interaction because there's a different pattern going on for the well the Blues and males than there are for the Reds who are females what about this one who thinks this is a significant interaction lots of shaking of heads very good because the lines are parallel it means the difference between men and women or difference between the lines is basically the same for the three alcohol conditions let's have a look at some bars significant interaction hands up no all right there's a few knows that's good this would be no interaction because again the the pattern across the bar is pretty much the same so his males down here females here and the pattern of like the basic the difference between the bar whites is pretty similar in the men and women so basically this pattern over here kind of looks the same as that pattern over there if you look at like the relative heights of the bars on the left they're pretty similar to the relative heights of the bars on the right what about this one I'll give you a clue we've seen this before earlier on in the lecture this again will be a significant interaction so we've got like no difference between the bars really here but then it suddenly becomes a difference there so the main thing to take home from this is really about trying to try to interpret interactions and we will we'll come back to this next week when we do repeats measures designs when we do three-way and over in the final week we'll also come back to interactions so you'll be getting lots of practice at it but yeah that's kind of main things bear in mind