Download Gender and Drink Type Interactions in Advertising: An Analysis using SPSS and more Lecture notes Design in PDF only on Docsity! DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 1 Chapter 15: Mixed design ANOVA Smart Alex’s Solutions Task 1 In the previous chapter we looked at an example in which participants viewed a total of nine mock adverts over three sessions. In these adverts there were three products (a brand of beer, Brain Death; a brand of wine, Dangleberry; and a brand of water, Puritan). These could be presented alongside positive, negative or neutral imagery. Over the three sessions and nine adverts, each type of product was paired with each type of imagery (read the previous chapter if you need more detail). After each advert participants rated the drinks on a scale ranging from −100 (dislike very much) through 0 (neutral) to 100 (like very much). The design had two repeated-‐measures independent variables: the type of drink (beer, wine or water) and the type of imagery used (positive, negative or neutral). Imagine that we also knew each participant’s gender. Men and women might respond differently to the products (because, in keeping with stereotypes, men might mostly drink lager whereas women might drink wine). Reanalyze the data, taking gender (a between-‐group variable) into account. The data are in the file MixedAttitude.sav. Run a three-‐way mixed ANOVA on these data. Running the analysis To carry out the analysis in SPSS, follow the same instructions that we did before. First of all, access the define factors dialog box by using the file path . We are using the same repeated-‐measures variables as in Chapter 13 of the book, so complete this dialog box exactly as shown there, and then click on to access the main dialog box. This box should be completed exactly as before, except that we must specify gender as a between-‐group variable by selecting it in the variables list and clicking to transfer it to the box labelled Between-‐Subjects Factors (Figure 1). DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 2 Figure 1 Gender has only two levels (male or female), so there is no need to specify contrasts for this variable; however, you should select simple contrasts for both drink and imagery. The addition of a between-‐group factor means that we can select post hoc tests for this variable by clicking on . This action brings up the post hoc test dialog box, which can be used as previously explained. However, we need not specify any post hoc tests here because the between-‐group factor has only two levels. The addition of an extra variable makes it necessary to choose a different graph than the one in the previous example. Click on to access the dialog box and place drink and imagery in the same slots as for the previous example, but also place gender in the slot labelled Separate Plots. When all three variables have been specified, don’t forget to click on to add this combination to the list of plots. By asking SPSS to plot the drink × imagery × gender interaction, we should get the same interaction graph as before, except that a separate version of this graph will be produced for male and female subjects. As far as other options are concerned, you should select the same ones that were chosen in Chapter 13. It is worth selecting estimated marginal means for all effects (because these values will help you to understand any significant effects), but to save space I did not ask for confidence intervals for these effects because we have considered this part of the output in some detail already. When all of the appropriate options have been selected, run the analysis. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 5 were previously present (in a balanced design, the inclusion of an extra variable should not affect these effects). By looking at the significance values it is clear that this prediction is true: there are still significant effects of the type of drink used, the type of imagery used, and the interaction of these two variables. In addition to the effects already described, we find that gender interacts significantly with the type of drink used (so, men and women respond differently to beer, wine and water regardless of the context of the advert). There is also a significant interaction of gender and imagery (so, men and women respond differently to positive, negative and neutral imagery regardless of the drink being advertised). Finally, the three-‐way interaction between gender, imagery and drink is significant, indicating that the way in which imagery affects responses to different types of drinks depends on whether the subject is male or female. The effects of the repeated-‐measures variables have been outlined in Chapter 13 and the pattern of these responses will not have changed, so, rather than repeat myself, I will concentrate on the new effects and the forgetful reader should look back at Chapter 13! Output 3 Tests of Within-Subjects Effects Measure: MEASURE_1 2092.344 2 1046.172 11.708 .000 2092.344 1.401 1493.568 11.708 .001 2092.344 1.567 1334.881 11.708 .000 2092.344 1.000 2092.344 11.708 .003 4569.011 2 2284.506 25.566 .000 4569.011 1.401 3261.475 25.566 .000 4569.011 1.567 2914.954 25.566 .000 4569.011 1.000 4569.011 25.566 .000 3216.867 36 89.357 3216.867 25.216 127.571 3216.867 28.214 114.017 3216.867 18.000 178.715 21628.678 2 10814.339 287.417 .000 21628.678 1.932 11196.937 287.417 .000 21628.678 2.000 10814.339 287.417 .000 21628.678 1.000 21628.678 287.417 .000 1998.344 2 999.172 26.555 .000 1998.344 1.932 1034.522 26.555 .000 1998.344 2.000 999.172 26.555 .000 1998.344 1.000 1998.344 26.555 .000 1354.533 36 37.626 1354.533 34.770 38.957 1354.533 36.000 37.626 1354.533 18.000 75.252 2624.422 4 656.106 19.593 .000 2624.422 3.251 807.186 19.593 .000 2624.422 4.000 656.106 19.593 .000 2624.422 1.000 2624.422 19.593 .000 495.689 4 123.922 3.701 .009 495.689 3.251 152.458 3.701 .014 495.689 4.000 123.922 3.701 .009 495.689 1.000 495.689 3.701 .070 2411.000 72 33.486 2411.000 58.524 41.197 2411.000 72.000 33.486 2411.000 18.000 133.944 Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Source DRINK DRINK * GENDER Error(DRINK) IMAGERY IMAGERY * GENDER Error(IMAGERY) DRINK * IMAGERY DRINK * IMAGERY * GENDER Error(DRINK*IMAGERY) Type III Sum of Squares df Mean Square F Sig. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 6 The effect of gender The main effect of gender is listed separately from the repeated-‐measures effects in a table labelled Tests of Between-‐Subjects Effects. Before looking at this table it is important to check the assumption of homogeneity of variance using Levene’s test (Output 4). SPSS produces a table listing Levene’s test for each of the repeated-‐measures variables in the data editor, and we need to look for any variable that has a significant value. The table showing Levene’s test indicates that variances are homogeneous for all levels of the repeated-‐measures variables (because all significance values are greater than .05). If any values were significant, then this would compromise the accuracy of the F-‐test for gender, and we would have to consider transforming all of our data to stabilize the variances between groups (one popular transformation is to take the square root of all values). Fortunately, in this example a transformation is unnecessary. Output 4 Output 5 Output 5 shows the ANOVA summary table for the main effect of gender, and (because the significance of .018 is less than the standard cut-‐off point of .05) we can report that there was a significant main effect of gender, F(1, 18) = 6.75, p < .05. This effect tells us that if we ignore all other variables, male subjects’ ratings were significantly different than females. If you requested that SPSS display means for the gender effect you should scan through your output and find the table in a section headed Estimated Marginal Means. The table of means for the main effect of gender with the associated standard errors is plotted alongside. It is clear from this graph that men’s ratings were generally significantly more positive than females. Levene's Test of Equality of Error Variancesa 1.009 1 18 .328 1.305 1 18 .268 1.813 1 18 .195 2.017 1 18 .173 1.048 1 18 .320 .071 1 18 .793 .317 1 18 .580 .804 1 18 .382 1.813 1 18 .195 Beer + Sexy Beer + Corpse Beer + Person in Armchair Wine + Sexy Wine + Corpse Wine + Person in Armchair Water + Sexy Water + Corpse Water + Person in Armchair F df1 df2 Sig. Tests the null hypothesis that the error variance of the dependent variable is equal across groups. Design: Intercept+GENDER - Within Subjects Design: DRINK+IMAGERY+DRINK*IMAGERY a. Tests of Between-Subjects Effects Measure: MEASURE_1 Transformed Variable: Average 1246.445 1 1246.445 144.593 .000 58.178 1 58.178 6.749 .018 155.167 18 8.620 Source Intercept GENDER Error Type III Sum of Squares df Mean Square F Sig. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 7 Therefore, men gave more positive ratings than women regardless of the drink being advertised and the type of imagery used in the advert. Figure 3 The interaction between gender and drink Gender interacted in some way with the type of drink used as a stimulus. Remembering that the effect of drink violated sphericity, we must report Greenhouse–Geisser-‐corrected values for this interaction with the between-‐group factor. From the summary table (Output 3) we should report that there was a significant interaction between the type of drink used and the gender of the subject, F(1.40, 25.22) = 25.57, p < .001. This effect tells us that the type of drink being advertised had a different effect on men and women. We can use the estimated marginal means to determine the nature of this interaction (or we could have asked SPSS for a plot of gender × drink). The means and interaction graph (Figure 4) show the meaning of this result. The graph shows the average male ratings of each drink ignoring the type of imagery with which it was presented (circles). The women’s scores are shown as squares. The graph clearly shows that male and female ratings are very similar for wine and water, but men seem to rate beer more highly than women — regardless of the type of imagery used. We could interpret this interaction as meaning that the type of drink being advertised influenced ratings differently in men and women. Specifically, ratings were similar for wine and water but males rated beer higher than women. This interaction can be clarified using the contrasts specified before the analysis. Figure 4 The interaction between gender and imagery Gender interacted in some way with the type of imagery used as a stimulus. The effect of imagery did not violate sphericity, so we can report the uncorrected F-‐value. From the Estimates Measure: MEASURE_1 9.600 .928 7.649 11.551 6.189 .928 4.238 8.140 Gender Male Female Mean Std. Error Lower Bound Upper Bound 95% Confidence Interval 9.60 6.19 0.00 5.00 10.00 15.00 Male Female 2. Gender * DRINK Measure: MEASURE_1 20.600 2.441 15.471 25.729 7.333 .765 5.726 8.940 .867 1.414 -2.103 3.836 3.067 2.441 -2.062 8.196 9.333 .765 7.726 10.940 6.167 1.414 3.197 9.136 DRINK 1 2 3 1 2 3 Gender Male Female Mean Std. Error Lower Bound Upper Bound 95% Confidence Interval 0 5 10 15 20 25 Beer Wine Water DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 10 Contrasts for repeated-measures variables We requested simple contrasts for the drink variable (for which water was used as the control category) and for the imagery category (for which neutral imagery was used as the control category). The table (Output 5) is the same as for the previous example except that the added effects of gender and its interaction with other variables are now included. So, for the main effect of drink, the first contrast compares level 1 (beer) against the base category (in this case, the last category, water); this result is significant, F(1, 18) = 15.37, p < .01. The next contrast compares level 2 (wine) with the base category (water) and confirms the significant difference found when gender was not included as a variable in the analysis, F(1, 18) = 19.92, p < .001. For the imagery main effect, the first contrast compares level 1 (positive) to the base category (neutral) and verifies the significant effect found by the post hoc tests, F(1, 18) = 134.87, p < .001. The second contrast confirms the significant difference found for the negative imagery condition compared to the neutral, F(1, 18) = 129.18, p < .001. No contrast was specified for gender. Output 5 Drink × gender interaction 1: beer vs. water, male vs. female The first interaction term looks at level 1 of drink (beer) compared to level 3 (water), comparing male and female scores. This contrast is highly significant, F(1, 18) = 28.97, p < .001. This result tells us that the increased ratings of beer compared to water found for men are not found for women. So, in the graph the squares representing female ratings of beer and water are roughly level; however, the circle representing male ratings of beer is much higher than the circle representing water. The positive contrast represents this difference, and so we can conclude that male ratings of beer (compared to water) were significantly greater than women’s ratings of beer (compared to water). Tests of Within-Subjects Contrasts Measure: MEASURE_1 1383.339 1 1383.339 15.371 .001 464.006 1 464.006 19.923 .000 2606.806 1 2606.806 28.965 .000 54.450 1 54.450 2.338 .144 1619.967 18 89.998 419.211 18 23.290 3520.089 1 3520.089 134.869 .000 3690.139 1 3690.139 129.179 .000 .556 1 .556 .021 .886 975.339 1 975.339 34.143 .000 469.800 18 26.100 514.189 18 28.566 320.000 1 320.000 1.686 .211 720.000 1 720.000 8.384 .010 36.450 1 36.450 .223 .642 2928.200 1 2928.200 31.698 .000 441.800 1 441.800 2.328 .144 480.200 1 480.200 5.592 .029 4.050 1 4.050 .025 .877 405.000 1 405.000 4.384 .051 3416.200 18 189.789 3416.200 18 189.789 1545.800 18 85.878 1662.800 18 92.378 IMAGERY Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 DRINK Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Level 1 vs. Level 3 Level 2 vs. Level 3 Source DRINK DRINK * GENDER Error(DRINK) IMAGERY IMAGERY * GENDER Error(IMAGERY) DRINK * IMAGERY DRINK * IMAGERY * GENDER Error(DRINK*IMAGERY) Type III Sum of Squares df Mean Square F Sig. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 11 Drink × gender interaction 2: wine vs. water, male vs. female The second interaction term compares level 2 of drink (wine) to level 3 (water), contrasting male and female scores. There is no significant difference for this contrast, F(1, 18) = 2.34, p = 0.14, which tells us that the difference between ratings of wine compared to water in males is roughly the same as in females. Therefore, overall, the drink × gender interaction has shown up a difference between males and females in how they rate beer (regardless of the type of imagery used). Imagery × gender interaction 1: positive vs. neutral, male vs. female The first interaction term looks at level 1 of imagery (positive) compared to level 3 (neutral), comparing male and female scores. This contrast is not significant (F < 1). This result tells us that ratings of drinks presented with positive imagery (relative to those presented with neutral imagery) were equivalent for males and females. This finding represents the fact that in the earlier graph of this interaction the squares and circles for both the positive and neutral conditions overlap (therefore male and female responses were the same). Imagery × gender interaction 2: negative vs. neutral, male vs. female The second interaction term looks at level 2 of imagery (negative) compared to level 3 (neutral), comparing male and female scores. This contrast is highly significant, F(1, 18) = 34.13, p < .001. This result tells us that the difference between ratings of drinks paired with negative imagery compared to neutral was different for men and women. Looking at the earlier graph of this interaction, this finding represents the fact that for men, ratings of drinks paired with negative imagery were relatively similar to ratings of drinks paired with neutral imagery (the circles have a fairly similar vertical position). However, if you look at the female ratings, then drinks were rated much less favourably when presented with negative imagery than when presented with neutral imagery (the square in the negative condition is much lower than the neutral condition). Therefore, overall, the imagery × gender interaction has shown up a difference between males and females in terms of their ratings of drinks presented with negative imagery compared to neutral; specifically, men seem less affected by negative imagery. Drink × imagery × gender interaction 1: beer vs. water, positive vs. neutral imagery, male vs. female The first interaction term compares level 1 of drink (beer) to level 3 (water), when positive imagery (level 1) is used compared to neutral (level 3) in males compared to females, F(1, 18) = 2.33, p = .144. The non-‐significance of this contrast tells us that the difference in ratings when positive imagery is used compared to neutral imagery is roughly equal when beer is used as a stimulus and when water is used, and these differences are equivalent in male and female subjects. In terms of the interaction graph it means that the distance between the circle and DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 12 the triangle in the beer condition is the same as the distance between the circle and the triangle in the water condition and that these distances are equivalent in men and women. Drink × imagery × gender interaction 2: beer vs. water, negative vs. neutral imagery, male vs. female The second interaction term looks at level 1 of drink (beer) compared to level 3 (water), when negative imagery (level 2) is used compared to neutral (level 3). This contrast is significant, F(1, 18) = 5.59, p < .05. This result tells us that the difference in ratings between beer and water when negative imagery is used (compared to neutral imagery) is different between men and women. If we plot ratings of beer and water across the negative and neutral conditions, for males (circles) and females (squares) separately, we see that ratings after negative imagery are always lower than ratings for neutral imagery except for men’s ratings of beer, which are actually higher after negative imagery. As such, this contrast tells us that the interaction effect reflects a difference in the way in which males rate beer compared to females when negative imagery is used compared to neutral. Males and females are similar in their pattern of ratings for water but different in the way in which they rate beer. Figure 7 Drink × imagery × gender interaction 3: wine vs. water, positive vs. neutral imagery, male vs. female. The third interaction term looks at level 2 of drink (wine) compared to level 3 (water), when positive imagery (level 1) is used compared to neutral (level 3) in males compared to females. This contrast is non-‐significant, F(1, 18) < 1. This result tells us that the difference in ratings when positive imagery is used compared to neutral imagery is roughly equal when wine is used as a stimulus and when water is used, and these differences are equivalent in male and female subjects. In terms of the interaction graph it means that the distance between the -20 -10 0 10 20 30 Negative Neutral Negative Neutral DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 15 Output 6 Output 6 shows the table of descriptive statistics from the two-‐way mixed ANOVA; the table has means at time 1 split according to whether the people were in the text messaging group or the control group, and then the means for the two groups at time 2. These means correspond to those plotted in Figure 9. Output 7 Output 8 We know that when we use repeated-‐measures we have to check the assumption of sphericity (Output 7). We also know that for independent designs we need to check the homogeneity of variance assumption. If the design is a mixed design then we have both repeated and independent measures, so we have to check both assumptions. In this case, we have only two levels of the repeated measure so the assumption of sphericity does not apply. Levene’s test produces a different test for each level of the repeated-‐measures variable (Output 8). In mixed designs, the homogeneity assumption has to hold for every level of the repeated-‐measures variable. At both levels of time, Levene’s test is non-‐significant (p = 0.77 before the experiment and p = .069 after the experiment). This means the assumption has not been broken at all (but it was quite close to being a problem after the experiment). Descriptive Statistics 64.8400 10.67973 25 65.6000 10.83590 25 65.2200 10.65467 50 52.9600 16.33116 25 61.8400 9.41046 25 57.4000 13.93278 50 Group Text Messagers Controls Total Text Messagers Controls Total Grammer at Time 1 Grammar at Time 2 Mean Std. Deviation N Mauchly's Test of Sphericityb Measure: MEASURE_1 1.000 .000 0 . 1.000 1.000 1.000 Within Subjects Effect TIME Mauchly's W Approx. Chi-Square df Sig. Greenhouse- Geisser Huynh-Feldt Lower-bound Epsilona Tests the null hypothesis that the error covariance matrix of the orthonormalized transformed dependent variables is proportional to an identity matrix. May be used to adjust the degrees of freedom for the averaged tests of significance. Corrected tests are displayed in the Tests of Within-Subjects Effects table. a. Design: Intercept+GROUP Within Subjects Design: TIME b. Levene's Test of Equality of Error Variancesa .089 1 48 .767 3.458 1 48 .069 Grammer at Time 1 Grammar at Time 2 F df1 df2 Sig. Tests the null hypothesis that the error variance of the dependent variable is equal across groups. Design: Intercept+GROUP Within Subjects Design: TIME a. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 16 Output 9 Output 10 Outputs 9 and 10 shows the main ANOVA summary tables. Like any two-‐way ANOVA, we still have three effects to find: two main effects (one for each independent variable) and one interaction term. The main effect of time is significant, so we can conclude that grammar scores were significantly affected by the time at which they were measured. The exact nature of this effect is easily determined because there were only two points in time (and so this main effect is comparing only two means). Figure 10 shows that grammar scores were higher before the experiment than after. So, before the experimental manipulation scores were higher than after, meaning that the manipulation had the net effect of significantly reducing grammar scores. This main effect seems rather interesting until you consider that these means include both text messagers and controls. There are three possible reasons for the drop in grammar scores: (1) the text messagers got worse and are dragging down the mean after the experiment; (2) the controls somehow got worse; or (3) the whole group just got worse and it had nothing to do with whether the children text-‐messaged or not. Until we examine the interaction, we won’t see which of these is true. Tests of Within-Subjects Effects Measure: MEASURE_1 1528.810 1 1528.810 15.457 .000 1528.810 1.000 1528.810 15.457 .000 1528.810 1.000 1528.810 15.457 .000 1528.810 1.000 1528.810 15.457 .000 412.090 1 412.090 4.166 .047 412.090 1.000 412.090 4.166 .047 412.090 1.000 412.090 4.166 .047 412.090 1.000 412.090 4.166 .047 4747.600 48 98.908 4747.600 48.000 98.908 4747.600 48.000 98.908 4747.600 48.000 98.908 Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Sphericity Assumed Greenhouse-Geisser Huynh-Feldt Lower-bound Source TIME TIME * GROUP Error(TIME) Type III Sum of Squares df Mean Square F Sig. Tests of Between-Subjects Effects Measure: MEASURE_1 Transformed Variable: Average 375891.610 1 375891.610 1933.002 .000 580.810 1 580.810 2.987 .090 9334.080 48 194.460 Source Intercept GROUP Error Type III Sum of Squares df Mean Square F Sig. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 17 Figure 10 The main effect of group is shown by the F-‐ratio in Output 10. The probability associated with this F-‐ratio is .09, which is just above the critical value of .05. Therefore, we must conclude that there was no significant main effect on grammar scores of whether children text-‐messaged or not. Again, this effect seems interesting enough, and mobile phone companies might certainly choose to cite it as evidence that text messaging does not affect your grammatical ability. However, remember that this main effect ignores the time at which grammatical ability is measured. It just means that if we took the average grammar score for text messagers (that’s including their score both before and after they started using their phone), and compared this to the mean of the controls (again including scores before and after) then these means would not be significantly different. The graph shows that when you ignore the time at which grammar was measured, the controls have slightly better grammar than the text messagers, but not significantly so. Main effects are not always that interesting and should certainly be viewed in the context of any interaction effects. The interaction effect in this example is shown by the F-‐ratio in the row labelled Time*Group, and because the probability of obtaining a value this big by chance is .047, which is just less than the criterion of .05, we can say that there is a significant interaction between the time at which grammar was measured and whether or not children were allowed to text-‐message within that time. The mean ratings in all conditions help us to interpret this effect. The significant interaction tells us that the change in grammar scores was significantly different in text messagers compared to controls. Looking at the interaction graph, we can see that although grammar scores fell in controls, the drop was much more marked in the text messagers; so, text messaging does seem to ruin your ability at grammar compared to controls.1 1 It’s interesting that the control group means dropped too. This could be because the control group were undisciplined and still used their mobile phones, or it could just be that the education system in this country is so underfunded that there is no one to teach English anymore! Group Before After M ea n G ra m m ar S co re (% ) 0 10 40 50 60 70 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 20 Figure 13: Error bar chart of mean personality disorder score before entering and after leaving the Big Brother house Output 11 Output 11 shows the table of descriptive statistics from the two-‐way mixed ANOVA; the table has mean borderline personality disorder (BPD) scores before entering the Big Brother house split according to whether the people were a contestant or not, and then the means for the two groups after leaving the house. These means correspond to those plotted in Figure 13. Output 12 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 21 Output 13 We know that when we use repeated-‐measures we have to check the assumption of sphericity (Output 12). However, we also know that for sphericity to be an issue we need at least three conditions. We have only two conditions here so sphericity does not need to be tested (and, therefore, SPSS produces a blank in the column labeled Sig.). We also need to check the homogeneity of variance assumption (Output 13). Levene’s test produces a different test for each level of the repeated-‐measures variable. In mixed designs, the homogeneity assumption has to hold for every level of the repeated-‐measures variable. At both levels of time, Levene’s test is non-‐significant (p = 0.061 before entering the Big Brother house and p = .088 after leaving). This means the assumption has not been significantly broken (but it was quite close to being a problem). Output 14 Output 15 Output 14 and 15 show the main ANOVA summary tables. Like any two-‐way ANOVA, we still have three effects to find: two main effects (one for each independent variable) and one interaction term. The main effect of time is not significant, so we can conclude that BPD scores were significantly affected by the time at which they were measured. The exact nature of this DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 22 effect is easily determined because there were only two points in time (and so this main effect is comparing only two means). Figure 14 shows that BPD scores were not significantly different after leaving the Big Brother house compared to before entering it. Figure 14 The main effect of group (bb) is shown by the F-‐ratio in Output 15. The probability associated with this F-‐ratio is .43, which is above the critical value of .05. Therefore, we must conclude that there was no significant main effect on BPD scores of whether the person was a Big Brother contestant or not. The graph shows that when you ignore the time at which BPD was measured, the contestants and controls are not significantly different. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 25 Figure 17 SPSS output Figure is a line graph of the angry pigs data. We can see that when participants played Tetris (blue line) in general their aggressive behaviour towards pigs decreased over time (except for between 6 months and 12 months when it actually increased slightly). However, when participants played Angry Birds, their aggressive behaviour towards pigs increased over time. Figure 18 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 26 Output 16 Output shows the estimated marginal means for the interaction between Tetris and time. These values correspond with those plotted in Figure . Output 17 Output shows the table of descriptive statistics from the two-‐way mixed ANOVA. Output 18 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 27 Output 19 We know that when we use repeated-‐measures we have to check the assumption of sphericity. We also know that for independent designs we need to check the homogeneity of variance assumption. If the design is a mixed design then we have both repeated and independent measures, so we have to check both assumptions. Output shows the results of Mauchly’s test for our repeated-‐measures variable Time. The value in the column labelled Sig is .170, which is larger than the cut off of .05, therefore it is non-‐significant and the assumption of sphericity has been met. Levene’s test produces a different test for each level of the repeated-‐measures variable. In mixed designs, the homogeneity assumption has to hold for every level of the repeated-‐measures variable. Output reveals that at each level of the variable Time, Levene’s test is significant (p < 0.05 in every case). This means the assumption has been broken. Output 20 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 30 Figure 19 Looking at Figure in comparison to Figure , we can see that aggressive behaviour in the real world was more erratic for the two video games than aggressive behaviour towards pigs. Figure shows that for Tetris, aggressive behaviour in the real world increased from time 1 (baseline) to time 3 (6 months) and then decreased from time 3 (6 months) to time 4 (12 months). For Angry Birds, on the other hand, aggressive behaviour in the real world initially increased from baseline to 1 month, it then decreased from 1 month to 6 months and then dramatically increased from 6 months to 12 months. The graph also shows that more aggressive behaviour was displayed when participants played Tetris compared to when they played Angry Birds at baseline, 1 month and 6 months; however, at 12 months participants engaged in more aggressive behaviour when playing Angry Birds compared to Tetris. Output 23 Output shows the results of Mauchly’s test for our repeated measures variable Time. The value in the column labelled Sig is .808 which is larger than the cut off of .05, therefore it is non-‐significant and the assumption of sphericity has been met. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 31 Output 24 Output 25 Output and Output show the main ANOVA summary tables. The main effect of Game was non-‐significant, indicating that (ignoring the time at which the aggression scores were measured), the type of game being played did not significantly affect participants’ aggression in the real world. The main effect of Time was also non-‐significant, so we can conclude that (ignoring the type of game being played), aggression was not significantly different at different points in time. The effect that we are most interested in is the Time × Game interaction, which was again non-‐significant. This effect tells us that change in aggression scores over time were not significantly different when participants played Tetris compared to when they played Angry Birds. Because none of the effects were significant it doesn’t make sense to conduct any contrasts. Therefore, we can conclude that playing Angry Birds does not make people more violent in general, just towards pigs. Task 6 My wife believes that she has received less friend requests from random men on Facebook since she changed her profile picture to a photo of us both. Imagine we took 40 women who had profiles on a social networking website; 17 of them had a relationship DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 32 status of ‘single’ and the remaining 23 had their status as ‘in a relationship’ (relationship_status). We asked these women to set their profile picture to a photo of them on their own (alone) and to count how many friend request they got from men over 3 weeks, then to switch it to a photo of them with a man (couple) and record their friend requests from random men over 3 weeks. The data are in the file ProfilePicture.sav. Run a mixed ANOVA to see if friend requests are affected by relationship status and type of profile picture. We need to run a 2 (relationship_status: single vs. in a relationship) × 2(photo: couple vs. alone) mixed ANOVA with repeated measures on the second variable. Your completed dialog box should look like Figure . Figure 20 Figure 21 DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 35 Output 28 Output and Output show the main ANOVA summary tables. Like any two-‐way ANOVA, we still have three effects to find: two main effects (one for each independent variable) and one interaction term. The main effect of relationship_status is significant, so we can conclude that, ignoring the type of profile picture, the number of friend requests was significantly affected by the relationship status of the woman. The exact nature of this effect is easily determined because there were only two levels of relationship status (and so this main effect is comparing only two means). Output 29 If we look at the estimated marginal means for relationship status in Output , we can see that the number of friend requests was significantly higher for single women (M = 5.94) compared to women who were in a relationship (M = 4.47). The main effect of Profile_Picture is shown by the F-‐ratio in Output . The probability associated with this F-‐ratio is shown as .000, which is smaller than the critical value of .05. Therefore, we can conclude that when ignoring relationship status, there was a significant main effect of whether the person was alone in their profile picture or with a partner on the number of friend requests. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 36 Output 30 Looking at the estimated marginal means for the profile picture variable (Output ), we can see that the number of friend requests was significantly higher when women were alone in their profile picture (M = 6.78) than when they were with a partner (M = 3.63). Note: we know that 1 = ‘in a couple’ and 2 = ‘alone’ in Output because this is how we coded the levels of the profile picture variable in the define dialog box in Figure . The interaction effect is the effect that we are most interested in and it is shown by the significance of the F-‐ratio in Output . We can see that the significance of the F-‐ratio is .010, which is less than the criterion of .05; therefore, we can say that there is a significant interaction between the relationship status of women and whether they had a photo of themselves alone or with a partner. The estimated marginal means in Output and the interaction graph (Figure ) help us to interpret this effect. The significant interaction seems to indicate that when displaying a photo of themselves alone rather than with a partner, the number of friend requests increases in both women in a relationship and single women. However, for single women this increase is greater than for women who are in a relationship. Output 31 Writing the results We can report the three effects from this analysis as follows: ! The main effect of relationship status was significant, F(1, 38) = 16.29, p < .001, indicating that single women received more friend requests than women who were in a relationship, regardless of their type of profile picture. ! The main effect of profile picture was significant, F(1, 38) = 114.77, p < .001, indicating that across all women, the number of friend requests was greater when displaying a photo alone rather than with a partner. DISCOVERING STATISTICS USING SPSS PROFESSOR ANDY P FIELD 37 ! The relationship status × profile picture interaction was significant, F(1, 38) = 7.41, p = .010, indicating that although number of friend requests increased in all women when they displayed a photo of themselves alone compared to when they displayed a photo of themselves with a partner, this increase was significantly greater for single women than for women who were in a relationship.