1. Study 1 - Body shape ideals
1.1 Correlations
corlist.s1 <- select(dazdat1, c(BMI, TVhrs, Age, daz_BMI, Discrep, daz_WHR, daz_WBR, daz_BHR))
apa.cor.table(corlist.s1, show.conf.interval=F, filename = "DAZcors_s1.doc")
## The ability to suppress reporting of reporting confidence intervals has been deprecated in this version.
## The function argument show.conf.interval will be removed in a later version.
##
##
## Means, standard deviations, and correlations with confidence intervals
##
##
## Variable M SD 1 2 3 4
## 1. BMI 24.44 4.83
##
## 2. TVhrs 10.98 10.62 .16
## [-.08, .38]
##
## 3. Age 19.68 4.96 .22 -.37**
## [-.02, .43] [-.55, -.15]
##
## 4. daz_BMI 20.76 3.01 .46** .01 .16
## [.25, .62] [-.22, .24] [-.07, .38]
##
## 5. Discrep 1.87 5.21 -.03 .15 -.11 -.52**
## [-.27, .21] [-.10, .37] [-.34, .13] [-.67, -.32]
##
## 6. daz_WHR 0.68 0.06 .00 -.25* .20 .19
## [-.23, .24] [-.45, -.01] [-.04, .41] [-.05, .40]
##
## 7. daz_WBR 0.78 0.05 .12 -.30* .31** .44**
## [-.12, .34] [-.50, -.07] [.08, .50] [.23, .61]
##
## 8. daz_BHR 0.87 0.05 -.13 .03 -.12 -.28*
## [-.35, .11] [-.21, .26] [-.34, .12] [-.48, -.05]
##
## 5 6 7
##
##
##
##
##
##
##
##
##
##
##
##
##
##
## -.13
## [-.35, .12]
##
## -.24 .76**
## [-.45, -.00] [.64, .85]
##
## .11 .56** -.11
## [-.13, .34] [.37, .70] [-.33, .13]
##
##
## Note. M and SD are used to represent mean and standard deviation, respectively.
## Values in square brackets indicate the 95% confidence interval.
## The confidence interval is a plausible range of population correlations
## that could have caused the sample correlation (Cumming, 2014).
## * indicates p < .05. ** indicates p < .01.
##
1.2 Differences between villages
Comparing body size preferences across villages, we see no significant differences. For body shape, women in Village 3 (low media Mestizo) created bodies with smaller hips compared to waists and busts. Women in Village 1 (high media Creole) also created bodies with smaller waists relative to busts compared to women in the other villages. Note Study 1 ANCOVA and correlation analyses reported in paper were run by TT in SPSS. R versions of analyses by LB are included here for completeness.
dazdat1$Village <- relevel(dazdat1$Village, ref = 2)
dazm1 <- lm(daz_BMI ~ Village + Age, data=dazdat1)
dazm2 <- lm(daz_WHR ~ Village + Age, data=dazdat1)
dazm3 <- lm(daz_WBR ~ Village + Age, data=dazdat1)
dazm4 <- lm(daz_BHR ~ Village + Age, data=dazdat1)
stargazer(dazm1,dazm2,dazm3,dazm4, type = "text")
##
## ==================================================================
## Dependent variable:
## ------------------------------------
## daz_BMI daz_WHR daz_WBR daz_BHR
## (1) (2) (3) (4)
## ------------------------------------------------------------------
## Village1 -1.803* 0.050*** 0.021 0.035**
## (0.935) (0.016) (0.015) (0.014)
##
## Village3 -1.686* -0.013 -0.030** 0.018
## (0.868) (0.015) (0.014) (0.013)
##
## Age 0.128 -0.001 0.001 -0.002*
## (0.087) (0.002) (0.001) (0.001)
##
## Constant 19.341*** 0.688*** 0.769*** 0.900***
## (1.734) (0.030) (0.028) (0.026)
##
## ------------------------------------------------------------------
## Observations 71 71 71 71
## R2 0.104 0.193 0.203 0.103
## Adjusted R2 0.064 0.156 0.167 0.063
## Residual Std. Error (df = 67) 2.917 0.051 0.047 0.044
## F Statistic (df = 3; 67) 2.594* 5.325*** 5.673*** 2.557*
## ==================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
village_proportions <- dazdat1 %>% group_by(Village) %>%
summarise('Ideal BMI' = mean(daz_BMI, na.rm=TRUE),
'Real BMI' = mean(BMI, na.rm=TRUE),
'Ideal WHR' = mean(daz_WHR, na.rm=TRUE),
'Real WHR' = mean(WHR, na.rm=TRUE),
'Ideal WBR' = mean(daz_WBR, na.rm=TRUE),
'Real WBR' = mean(WBR, na.rm=TRUE),
'Ideal BHR' = mean(daz_BHR, na.rm=TRUE),
'Real BHR' = mean(BHR, na.rm=TRUE))
as.data.frame(village_proportions)
#make data long and do violin plots on proportions
f1.1 <- dazdat1 %>% select(ID, Village,WHR, BHR, WBR, daz_WHR, daz_WBR, daz_BHR) %>%
rename(Real_WHR = WHR,
Real_WBR = WBR,
Real_BHR = BHR) %>%
pivot_longer(3:8, values_to = "Ratio") %>%
separate(name, c("Type", "Dimension"),sep="_") %>%
ggplot(aes(Village, Ratio,fill = Type)) + geom_violin() + facet_wrap("Dimension") +
scale_fill_brewer() + theme_minimal() + theme(legend.position="none")
f1.2 <- dazdat1 %>% select(ID, Village, BMI, daz_BMI) %>%
rename(Real_BMI = BMI) %>%
pivot_longer(3:4, values_to="BMI") %>%
separate(name, c("Type", "Dimension"),sep="_")%>%
ggplot(aes(Village, BMI,fill = Type)) + geom_violin() +
scale_fill_brewer(labels = c("Ideal","Real")) + theme_minimal()
Figure1 <- cowplot::plot_grid(f1.1, f1.2, ncol = 1)
cowplot::save_plot(Figure1, file="Figure1.jpg")
Figure1
1.3 Regressions controlling for age
Despite zero order correlations above, once we control for age, overall TV consumption no longer relates to body shape variables although reported frequency of US tv consumption does.
#checking all TV hours associations
dazmtv1 <- lm(daz_BMI ~ TVhrs + Age, data=dazdat1)
dazmtv2 <- lm(daz_WHR ~ TVhrs + Age, data=dazdat1)
dazmtv3 <- lm(daz_WBR ~ TVhrs + Age, data=dazdat1)
dazmtv4 <- lm(daz_BHR ~ TVhrs + Age, data=dazdat1)
stargazer(dazmtv1,dazmtv2,dazmtv3,dazmtv4, type = "text")
##
## ==================================================================
## Dependent variable:
## ------------------------------------
## daz_BMI daz_WHR daz_WBR daz_BHR
## (1) (2) (3) (4)
## ------------------------------------------------------------------
## TVhrs 0.023 -0.001 -0.001* -0.0001
## (0.036) (0.001) (0.001) (0.001)
##
## Age 0.118 0.001 0.002* -0.001
## (0.078) (0.001) (0.001) (0.001)
##
## Constant 18.172*** 0.666*** 0.749*** 0.896***
## (1.759) (0.032) (0.029) (0.027)
##
## ------------------------------------------------------------------
## Observations 71 71 71 71
## R2 0.033 0.074 0.134 0.015
## Adjusted R2 0.004 0.046 0.108 -0.014
## Residual Std. Error (df = 68) 3.008 0.054 0.049 0.046
## F Statistic (df = 2; 68) 1.157 2.703* 5.249*** 0.518
## ==================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
#checking correlations which were sig with other TV variables
dazmtv2am <- lm(daz_WHR ~ TV_Am + Age, data=dazdat1)
dazmtv3am <- lm(daz_WBR ~ TV_Am + Age, data=dazdat1)
dazmtv1sp <- lm(daz_WBR ~ TV_Sp + Age, data=dazdat1)
dazmtv2fms <- lm(daz_WBR ~ FILM_Sp + Age, data=dazdat1)
dazmtv3fma <- lm(daz_WBR ~ FILM_Am + Age, data=dazdat1)
stargazer(dazmtv2am,dazmtv3am,dazmtv1sp,dazmtv2fms,dazmtv3fma, type = "text")
##
## ==========================================================================
## Dependent variable:
## --------------------------------------------
## daz_WHR daz_WBR
## (1) (2) (3) (4) (5)
## --------------------------------------------------------------------------
## TV_Am -0.013** -0.011**
## (0.005) (0.005)
##
## TV_Sp -0.005
## (0.005)
##
## FILM_Sp -0.009*
## (0.005)
##
## FILM_Am -0.009*
## (0.005)
##
## Age 0.0004 0.002 0.003** 0.002* 0.002*
## (0.001) (0.001) (0.001) (0.001) (0.001)
##
## Constant 0.699*** 0.772*** 0.747*** 0.754*** 0.755***
## (0.036) (0.033) (0.035) (0.031) (0.031)
##
## --------------------------------------------------------------------------
## Observations 71 71 71 71 71
## R2 0.121 0.159 0.109 0.131 0.133
## Adjusted R2 0.095 0.134 0.082 0.106 0.108
## Residual Std. Error (df = 68) 0.053 0.048 0.050 0.049 0.049
## F Statistic (df = 2; 68) 4.664** 6.438*** 4.145** 5.138*** 5.219***
## ==========================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
1.4 Comparing real and ideal body size/shape
Reviewer 1 requested that we compare real and ideal body dimensions. We find that women’s preferred size and shape differs significantly from actual weigh and shape acros all calculated proportions.
t.test(dazdat1$BMI, dazdat1$daz_BMI, paired = TRUE, alternative = "two.sided")
##
## Paired t-test
##
## data: dazdat1$BMI and dazdat1$daz_BMI
## t = 7.0919, df = 70, p-value = 8.49e-10
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 2.648874 4.721669
## sample estimates:
## mean of the differences
## 3.685271
t.test(dazdat1$WHR, dazdat1$daz_WHR, paired = TRUE, alternative = "two.sided")
##
## Paired t-test
##
## data: dazdat1$WHR and dazdat1$daz_WHR
## t = 17.871, df = 70, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 0.1243026 0.1555320
## sample estimates:
## mean of the differences
## 0.1399173
t.test(dazdat1$WBR, dazdat1$daz_WBR, paired = TRUE, alternative = "two.sided")
##
## Paired t-test
##
## data: dazdat1$WBR and dazdat1$daz_WBR
## t = 4.8623, df = 70, p-value = 6.876e-06
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 0.02141300 0.05119617
## sample estimates:
## mean of the differences
## 0.03630458
t.test(dazdat1$BHR, dazdat1$daz_BHR, paired = TRUE, alternative = "two.sided")
##
## Paired t-test
##
## data: dazdat1$BHR and dazdat1$daz_BHR
## t = 11.105, df = 70, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 0.07104182 0.10214756
## sample estimates:
## mean of the differences
## 0.08659469
2. Study 2 - Body image and eating attitudes
2.1 Data correlations
corlist <- select(dazdat1, c(BMI, BAStotal, BSQtotal, SATAQtotal, SATAQinternalisation, SATAQpressure, SATAQinformation, EATtotal, TVhrs, Age, Discrep,daz_BMI))
apa.cor.table(corlist, show.conf.interval=F, filename = "DAZcors1.doc")
## The ability to suppress reporting of reporting confidence intervals has been deprecated in this version.
## The function argument show.conf.interval will be removed in a later version.
##
##
## Means, standard deviations, and correlations with confidence intervals
##
##
## Variable M SD 1 2 3
## 1. BMI 24.44 4.83
##
## 2. BAStotal 59.42 6.23 .02
## [-.21, .25]
##
## 3. BSQtotal 15.59 7.81 .48** -.42**
## [.28, .64] [-.59, -.20]
##
## 4. SATAQtotal 46.54 18.71 .17 -.40** .41**
## [-.07, .39] [-.58, -.18] [.20, .59]
##
## 5. SATAQinternalisation 12.13 6.24 .08 -.34** .38**
## [-.16, .31] [-.53, -.11] [.16, .56]
##
## 6. SATAQpressure 13.29 7.42 .22 -.41** .46**
## [-.02, .43] [-.59, -.20] [.26, .63]
##
## 7. SATAQinformation 13.37 4.21 .10 -.25* .23
## [-.14, .33] [-.46, -.02] [-.00, .44]
##
## 8. EATtotal 5.40 5.90 .40** -.29* .54**
## [.18, .58] [-.49, -.06] [.35, .69]
##
## 9. TVhrs 10.98 10.62 .16 -.22 .29*
## [-.08, .38] [-.43, .01] [.07, .49]
##
## 10. Age 19.68 4.96 .22 .30* -.11
## [-.02, .43] [.07, .50] [-.34, .12]
##
## 11. Discrep 1.87 5.21 -.03 -.26* .13
## [-.27, .21] [-.47, -.03] [-.11, .36]
##
## 12. daz_BMI 20.76 3.01 .46** -.05 .15
## [.25, .62] [-.28, .18] [-.09, .37]
##
## 4 5 6 7 8 9
##
##
##
##
##
##
##
##
##
##
##
## .87**
## [.80, .92]
##
## .91** .74**
## [.85, .94] [.61, .83]
##
## .77** .53** .54**
## [.65, .85] [.34, .68] [.35, .69]
##
## .33** .33** .39** .20
## [.10, .52] [.10, .52] [.17, .57] [-.03, .42]
##
## .10 .07 .06 .23 .21
## [-.14, .33] [-.17, .30] [-.18, .29] [-.01, .44] [-.03, .42]
##
## -.34** -.19 -.24* -.47** -.03 -.37**
## [-.53, -.11] [-.40, .05] [-.45, -.00] [-.64, -.26] [-.26, .21] [-.55, -.15]
##
## .09 .01 .07 .11 .06 .15
## [-.15, .33] [-.23, .25] [-.17, .30] [-.13, .34] [-.18, .30] [-.10, .37]
##
## .11 .09 .10 .06 .05 .01
## [-.13, .33] [-.15, .32] [-.14, .32] [-.18, .29] [-.19, .28] [-.22, .24]
##
## 10 11
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
## -.11
## [-.34, .13]
##
## .16 -.52**
## [-.07, .38] [-.67, -.32]
##
##
## Note. M and SD are used to represent mean and standard deviation, respectively.
## Values in square brackets indicate the 95% confidence interval.
## The confidence interval is a plausible range of population correlations
## that could have caused the sample correlation (Cumming, 2014).
## * indicates p < .05. ** indicates p < .01.
##
2.2 Comparing villages on body image traits
As before, published analyses were run in SPSS but we include linear models for village for completeness. We can see that women in Village 3 (Mestizo, low media) have lower body dissatsifaction than women in the other villages. Women in Village 2 (Mestizo high media) have higher levels of thin ideal internalisation than other women. Older women also show less thin ideal internalisation.
mv1 <- lm(BAStotal ~ Age + Village, data=dazdat1)
mv2 <- lm(BSQtotal ~ Age + Village, data=dazdat1)
mv3 <- lm(SATAQtotal ~ Age + Village, data=dazdat1)
mv4 <- lm(EATtotal ~ Age + Village, data=dazdat1)
stargazer(mv1,mv2,mv3,mv4, type="text")
##
## ====================================================================================================
## Dependent variable:
## --------------------------------------------------------------------------------
## BAStotal BSQtotal SATAQtotal EATtotal
## (1) (2) (3) (4)
## ----------------------------------------------------------------------------------------------------
## Age 0.316* 0.169 -1.132** 0.158
## (0.180) (0.225) (0.485) (0.173)
##
## Village1 1.967 -5.854** -15.121*** -4.320**
## (1.931) (2.416) (5.257) (1.878)
##
## Village3 1.152 0.852 -16.103*** -0.930
## (1.791) (2.241) (4.833) (1.727)
##
## Constant 52.209*** 13.891*** 78.562*** 3.960
## (3.581) (4.479) (9.661) (3.452)
##
## ----------------------------------------------------------------------------------------------------
## Observations 71 71 70 70
## R2 0.105 0.109 0.279 0.075
## Adjusted R2 0.064 0.069 0.246 0.033
## Residual Std. Error 6.022 (df = 67) 7.533 (df = 67) 16.247 (df = 66) 5.805 (df = 66)
## F Statistic 2.607* (df = 3; 67) 2.720* (df = 3; 67) 8.516*** (df = 3; 66) 1.787 (df = 3; 66)
## ====================================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
2.3 Testing the sociocultural model.
The model predicts that BMI and internalisation should predict body dissatisfaction and dissatisfaction should in turn predict EAT scores. The relationship between BMI/internalisation and EAT scores should become ns once BSQ is included. This is what we find in regression models. The published paper focuses on mediation models below, but we have the regression models here for completeness and to show that including age and BMI does not change result for mediation models below.
#create complete dataset for hierarchical models
dazdat1comp <- subset(dazdat1, EATtotal>-1)
m0 <- lm(BSQtotal~ Age + BMI + SATAQinternalisation, data=dazdat1comp)
m1 <- lm(EATtotal~Age + BMI , data=dazdat1comp)
m2<- lm(EATtotal~Age + BMI + SATAQinternalisation, data=dazdat1comp)
m3<- lm(EATtotal~Age + BMI + SATAQinternalisation+BSQtotal, data=dazdat1comp)
#apa.reg.table(m1,m2,m3,m4)
stargazer(m0,m1,m2,m3, type="text")
##
## =============================================================================================================
## Dependent variable:
## ----------------------------------------------------------------------------------------
## BSQtotal EATtotal
## (1) (2) (3) (4)
## -------------------------------------------------------------------------------------------------------------
## Age -0.244 -0.141 -0.068 0.001
## (0.161) (0.136) (0.133) (0.128)
##
## BMI 0.794*** 0.514*** 0.470*** 0.243
## (0.162) (0.138) (0.134) (0.148)
##
## SATAQinternalisation 0.389*** 0.271** 0.160
## (0.126) (0.104) (0.105)
##
## BSQtotal 0.286***
## (0.096)
##
## Constant -3.687 -4.392 -8.039** -6.985*
## (4.808) (3.868) (3.962) (3.761)
##
## -------------------------------------------------------------------------------------------------------------
## Observations 70 70 70 70
## R2 0.372 0.171 0.249 0.340
## Adjusted R2 0.343 0.146 0.215 0.299
## Residual Std. Error 6.347 (df = 66) 5.454 (df = 67) 5.230 (df = 66) 4.943 (df = 65)
## F Statistic 13.028*** (df = 3; 66) 6.921*** (df = 2; 67) 7.305*** (df = 3; 66) 8.355*** (df = 4; 65)
## =============================================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
2.3.2 Exploratory models on the role of BAS
We can also use BAS and see if positive body image shows the same (inverse) pattern as body disatisfaction measured by the BSQ - but we find that there’s no effect of BAS on EAT once the direct effect of internalisation is accounted for.
# Substituting BAS instead of BSQ
labels=list(X="SATAQinternalisation", M="BAStotal", Y="EATtotal")
process.model <- tripleEquation(X="SATAQinternalisation", M="BAStotal", Y="EATtotal")
fit4 <- sem(process.model, data = dazdat1)
summary(fit4)
## lavaan 0.6-8 ended normally after 16 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 5
##
## Used Total
## Number of observations 70 71
##
## Model Test User Model:
##
## Test statistic 0.000
## Degrees of freedom 0
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Expected
## Information saturated (h1) model Structured
##
## Regressions:
## Estimate Std.Err z-value P(>|z|)
## BAStotal ~
## SATAQntrnl (a) -0.341 0.113 -3.022 0.003
## EATtotal ~
## SATAQntrnl (c) 0.245 0.111 2.201 0.028
## BAStotal (b) -0.190 0.111 -1.717 0.086
##
## Variances:
## Estimate Std.Err z-value P(>|z|)
## .BAStotal 34.257 5.790 5.916 0.000
## .EATtotal 29.429 4.974 5.916 0.000
##
## Defined Parameters:
## Estimate Std.Err z-value P(>|z|)
## indirect 0.065 0.043 1.493 0.135
## direct 0.245 0.111 2.201 0.028
## total 0.310 0.107 2.900 0.004
## prop.mediated 0.209 0.151 1.384 0.166
statisticalDiagram(4, labels=labels,fit=fit4, whatLabel="est")
labels=list(X="BAStotal", M="BSQtotal", Y="EATtotal")
process.model <- tripleEquation(X="BAStotal", M="BSQtotal", Y="EATtotal")
fit5 <- sem(process.model, data = dazdat1)
summary(fit5)
## lavaan 0.6-8 ended normally after 13 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 5
##
## Used Total
## Number of observations 70 71
##
## Model Test User Model:
##
## Test statistic 0.000
## Degrees of freedom 0
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Expected
## Information saturated (h1) model Structured
##
## Regressions:
## Estimate Std.Err z-value P(>|z|)
## BSQtotal ~
## BAStotal (a) -0.518 0.136 -3.810 0.000
## EATtotal ~
## BAStotal (c) -0.075 0.104 -0.726 0.468
## BSQtotal (b) 0.382 0.083 4.600 0.000
##
## Variances:
## Estimate Std.Err z-value P(>|z|)
## .BSQtotal 50.093 8.467 5.916 0.000
## .EATtotal 24.164 4.084 5.916 0.000
##
## Defined Parameters:
## Estimate Std.Err z-value P(>|z|)
## indirect -0.198 0.067 -2.934 0.003
## direct -0.075 0.104 -0.726 0.468
## total -0.273 0.108 -2.534 0.011
## prop.mediated 0.724 0.300 2.411 0.016
statisticalDiagram(4, labels=labels,fit=fit5, whatLabel="est")
labels=list(X="BAStotal", M="SATAQinternalisation", Y="BSQtotal")
process.model <- tripleEquation(X="BAStotal", M="SATAQinternalisation", Y="BSQtotal")
fit6 <- sem(process.model, data = dazdat1)
summary(fit6)
## lavaan 0.6-8 ended normally after 16 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 5
##
## Used Total
## Number of observations 70 71
##
## Model Test User Model:
##
## Test statistic 0.000
## Degrees of freedom 0
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Expected
## Information saturated (h1) model Structured
##
## Regressions:
## Estimate Std.Err z-value P(>|z|)
## SATAQinternalisation ~
## BAStotal (a) -0.338 0.112 -3.022 0.003
## BSQtotal ~
## BAStotal (c) -0.404 0.139 -2.910 0.004
## SATAQntrnl (b) 0.336 0.140 2.408 0.016
##
## Variances:
## Estimate Std.Err z-value P(>|z|)
## .SATAQintrnlstn 33.942 5.737 5.916 0.000
## .BSQtotal 46.261 7.820 5.916 0.000
##
## Defined Parameters:
## Estimate Std.Err z-value P(>|z|)
## indirect -0.114 0.060 -1.883 0.060
## direct -0.404 0.139 -2.910 0.004
## total -0.518 0.136 -3.810 0.000
## prop.mediated 0.219 0.121 1.817 0.069
statisticalDiagram(4, labels=labels,fit=fit6, whatLabel="est")
labels=list(X="TVhrs", M="SATAQinternalisation", Y="BSQtotal")
process.model <- tripleEquation(X="TVhrs", M="SATAQinternalisation", Y="BSQtotal")
fit7 <- sem(process.model, data = dazdat1)
summary(fit7)
## lavaan 0.6-8 ended normally after 15 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 5
##
## Used Total
## Number of observations 70 71
##
## Model Test User Model:
##
## Test statistic 0.000
## Degrees of freedom 0
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Expected
## Information saturated (h1) model Structured
##
## Regressions:
## Estimate Std.Err z-value P(>|z|)
## SATAQinternalisation ~
## TVhrs (a) 0.040 0.070 0.573 0.566
## BSQtotal ~
## TVhrs (c) 0.193 0.079 2.461 0.014
## SATAQntrnl (b) 0.451 0.134 3.379 0.001
##
## Variances:
## Estimate Std.Err z-value P(>|z|)
## .SATAQintrnlstn 38.190 6.455 5.916 0.000
## .BSQtotal 47.728 8.068 5.916 0.000
##
## Defined Parameters:
## Estimate Std.Err z-value P(>|z|)
## indirect 0.018 0.032 0.565 0.572
## direct 0.193 0.079 2.461 0.014
## total 0.212 0.085 2.502 0.012
## prop.mediated 0.086 0.143 0.601 0.548
statisticalDiagram(4, labels=labels,fit=fit7, whatLabel="est")
2.4 SEM with lavaan to investigate the possibility that BAS / SATAQ -> BSQ -> EAT
model1 <- ' # direct effects
EATtotal ~ d*BAStotal + e*SATAQinternalisation
# mediator
BSQtotal ~ a*BAStotal + b*SATAQinternalisation
EATtotal ~ c*BSQtotal
# indirect effect (a*b)
abc := a*b*c
# total effect
total := d*e + (a*b*c)
'
fit1 <- sem(model1, data = dazdat1)
summary(fit1)
## lavaan 0.6-8 ended normally after 14 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 7
##
## Used Total
## Number of observations 70 71
##
## Model Test User Model:
##
## Test statistic 0.000
## Degrees of freedom 0
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Expected
## Information saturated (h1) model Structured
##
## Regressions:
## Estimate Std.Err z-value P(>|z|)
## EATtotal ~
## BAStotal (d) -0.047 0.105 -0.452 0.652
## SATAQntrnl (e) 0.126 0.104 1.217 0.224
## BSQtotal ~
## BAStotal (a) -0.404 0.139 -2.910 0.004
## SATAQntrnl (b) 0.336 0.140 2.408 0.016
## EATtotal ~
## BSQtotal (c) 0.353 0.085 4.130 0.000
##
## Variances:
## Estimate Std.Err z-value P(>|z|)
## .EATtotal 23.663 4.000 5.916 0.000
## .BSQtotal 46.261 7.820 5.916 0.000
##
## Defined Parameters:
## Estimate Std.Err z-value P(>|z|)
## abc -0.048 0.024 -1.991 0.046
## total -0.054 0.025 -2.168 0.030
AIC(fit1)
## [1] 901.1773
lavaanPlot(model = fit1, node_options = list(shape = "box", fontname =
"Helvetica"), edge_options = list(color = "grey"), coefs = TRUE,covs=
TRUE,stars = c("regress"))
3. Study 3 - Longitudinal sociocultural analyses
3.1 Main analyses
#center variables
# library(robumeta)
# dazdat$SATAQintern_cent <- group.center(dazdat$SATAQinternalisation, dazdat$ID)
# dazdat$BSQtotal_cent <- group.center(dazdat$BSQtotal, dazdat$ID)
#show how many ppl participated more than once
dazdat1 %>% group_by(times_participated,Village) %>%
count()
#select repeat participants only
dazdatlong <- subset(dazdat, times_participated>1)
#run sequential models for paths in sociocultural model
ml0 <- lmer(BSQtotal~SATAQinternalisation + (1|ID), data=dazdatlong)
ml2<- lmer(EATtotal~SATAQinternalisation+ (1|ID), data=dazdatlong)
ml3<- lmer(EATtotal~SATAQinternalisation+BSQtotal+ (1|ID), data=dazdatlong)
ml4<- lmer(EATtotal~SATAQinternalisation+BSQtotal + TIME + (1|ID), data=dazdatlong)
stargazer(ml0, ml2,ml3,ml4, type="text")
##
## =========================================================
## Dependent variable:
## ------------------------------------
## BSQtotal EATtotal
## (1) (2) (3) (4)
## ---------------------------------------------------------
## SATAQinternalisation 0.407*** 0.236* 0.054 0.050
## (0.141) (0.127) (0.117) (0.112)
##
## BSQtotal 0.449*** 0.428***
## (0.092) (0.088)
##
## TIMET2 -0.077
## (1.720)
##
## TIMET3 3.443***
## (1.263)
##
## Constant 11.502*** 3.842** -1.328 -2.354
## (1.821) (1.656) (1.791) (1.867)
##
## ---------------------------------------------------------
## Observations 76 76 76 76
## Log Likelihood -258.344 -250.923 -241.948 -235.370
## Akaike Inf. Crit. 524.687 509.847 493.896 484.740
## Bayesian Inf. Crit. 534.010 519.170 505.550 501.055
## =========================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
3.2 Checking basic effects of timepoint
For completeness, we see that Timepoint 3 was also significantly different from Timepoint 1 even without BSQ score and SATAQ in the model.
#basic effects of timepoint
mlt1<- lmer(SATAQinternalisation~ TIME + (1|ID), data=dazdatlong)
mlt2<- lmer(BSQtotal~TIME + (1|ID), data=dazdatlong)
mlt3<- lmer(EATtotal~TIME + (1|ID), data=dazdatlong)
stargazer(mlt1,mlt2,mlt3, type="text")
##
## ===========================================================
## Dependent variable:
## ---------------------------------------
## SATAQinternalisation BSQtotal EATtotal
## (1) (2) (3)
## -----------------------------------------------------------
## TIMET2 -0.771 -1.729 -0.839
## (1.860) (2.346) (1.965)
##
## TIMET3 0.598 0.867 3.857***
## (1.366) (1.717) (1.438)
##
## Constant 11.020*** 15.949*** 5.007***
## (1.088) (1.386) (1.161)
##
## -----------------------------------------------------------
## Observations 76 76 76
## Log Likelihood -239.821 -257.299 -244.367
## Akaike Inf. Crit. 489.641 524.598 498.735
## Bayesian Inf. Crit. 501.295 536.252 510.388
## ===========================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
3.3 Repeating key models with age added
Reivewer 1 asked if age was important in these analyses. Note age is subsumed entirely within the participant random effect and timepoint fixed effect. We therefore expect suppression effects with timepoint in particular. Note that with age added to final model, neither timepoint nor age is significant.
mlt1a<- lmer(SATAQinternalisation~ TIME + Age + (1|ID), data=dazdatlong)
mlt2a<- lmer(BSQtotal~TIME + Age + (1|ID), data=dazdatlong)
mlt3a<- lmer(EATtotal~TIME + Age + (1|ID), data=dazdatlong)
stargazer(mlt1a,mlt2a,mlt3a, type="text")
##
## ==========================================================
## Dependent variable:
## --------------------------------------
## SATAQinternalisation BSQtotal EATtotal
## (1) (2) (3)
## ----------------------------------------------------------
## TIMET2 -1.423 -2.876 -1.965
## (1.916) (2.410) (2.033)
##
## TIMET3 -0.410 -1.081 1.965
## (1.588) (1.994) (1.672)
##
## Age 0.285 0.546* 0.522**
## (0.234) (0.291) (0.234)
##
## Constant 5.673 5.699 -4.772
## (4.523) (5.633) (4.536)
##
## ----------------------------------------------------------
## Observations 76 76 76
## Log Likelihood -239.618 -255.889 -242.601
## Akaike Inf. Crit. 491.235 523.777 497.201
## Bayesian Inf. Crit. 505.219 537.762 511.186
## ==========================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
ml0a <- lmer(BSQtotal~SATAQinternalisation + Age + (1|ID), data=dazdatlong)
ml4a<- lmer(EATtotal~SATAQinternalisation+BSQtotal + TIME +Age + (1|ID), data=dazdatlong)
stargazer(ml0a, ml4a, type="text")
##
## =================================================
## Dependent variable:
## ----------------------------
## BSQtotal EATtotal
## (1) (2)
## -------------------------------------------------
## SATAQinternalisation 0.378*** 0.032
## (0.141) (0.112)
##
## BSQtotal 0.407***
## (0.090)
##
## TIMET2 -0.800
## (1.803)
##
## TIMET3 2.421
## (1.472)
##
## Age 0.366 0.293
## (0.232) (0.211)
##
## Constant 4.280 -7.310*
## (4.888) (4.047)
##
## -------------------------------------------------
## Observations 76 76
## Log Likelihood -257.654 -235.087
## Akaike Inf. Crit. 525.308 486.174
## Bayesian Inf. Crit. 536.962 504.820
## =================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
