Correlates of BMI were examined at T1. Age was correlated with it so it was controlled at step 1 of the analysis. 23.3% of variance in BMI was predicted by age at step 1 (F1, 159=48.26, p<.001). At step 2, covariates predicted an additional 4.1% of its variance increasing it to 27.4% (R2change: F4, 156=14.7, p<.001). At step 3, an additional 1.9% of its variance was predicted by sleep parameters increasing it to 29.3%, R2change: F5, 155=12.82, p<.001. At step 3, high BMI was related to more SD (p<.05) after controlling demographics and covariates (i.e. strength/flexibility, nocturnal TV & device use), see Table 2.
Risk factors for high BMI were examined at T2. T1 BMI and age were correlated with it at T2 and were entered at step 1 of the analysis. At step 1, they predicted 98.1% of the variance in BMI (F2, 152=3950, p<.001). At step 2 (F5, 149=1554 p<.001) and step 3 (F6, 148=1287, p<.001), the IVs and covariates did not increase its explained variance (R2=98.1%) and so no IVs predicted it after controlling T1 BMI, age and covariates (strength/flexibility, nocturnal TV & device use), see Table 3.
Correlates of WHR were examined at T1. Age and gender were correlated with it and so were entered at step 1, covariates correlated with it were entered at step 2 and IVs were entered at step 3. Results indicated that 41.3% of variance in WHR was predicted at step 1 (F2, 158= 55.7, p<.001). At step 2, covariates predicted an additional 1% of its variance increasing it to 42.3% (R2change: F4, 156=28.6, p<.001). At step 3, an additional 0.9% of its variance was predicted increasing it to 43.2%, R2change: F6, 154=19.5, p<.001. WHR was related to male gender and older age (p-value <.005), but not measures of sleep, see Table 4.
Risk factors for high WHR were examined at T2. WHR T1, age and gender were entered at step 1, covariates at step 2 and IVs at step 3, if they were correlated with WHR at T2. At step 1, they predicted 81.1% of its variance (F3, 151 =216.5, p<.001). At step 2, covariates predicted an additional 0.1% of its variance increasing it to 81.2% (R2change: F4, 150=162.2, p<.001). At step 3, an additional 0.1% of its variance was predicted increasing it to 81.3%, R2change: F5, 149=130.0, p<.001. Higher WHR at T2 was predicted by male gender and older age (p<.005) after controlling for T1 WHR, see Table 5.
Binary logistic regression examined predictors of obesity category (i.e. overweight/obesity vs. normal weight) at T1. Age was included at step 1, covariates were entered at step 2 and IVs were entered at step 3 of the analysis. At step 1, age predicted 17% (Cox & Snell R2) of its variance, χ²(1) (N=161)=30.07, p<.001; model correctly classified 73.3% of cases. At step 2, covariates predicted an additional 1% of its variance increasing it to 18%, χ²(4) (N=161)= 31.9, p<.001; model correctly classified 73.9% of cases. At step 3 (χ²(6) (N=161)=47.3, p<.001), IVs predicted an additional 7.4% of its variance increasing it to 25.4%; model correctly classified 73.9% of cases. Only longer awake time during sleep was related to obesity category after controlling age and the covariates, see Table 6. Older age was related to 11% increased odds of being overweight/obese whereas awake time was linked to 2% increased odds of overweight/obesity.
Binary logistic regression analysis examined risk factors for obesity category at T2. T1 obesity category and age were entered at step 1, covariates were entered at step 2 and IVs correlated with it were entered at step 3 of the analysis. At step 1, the factors predicted 71.2% (Cox & Snell R2) of its variance, χ²(2) (N=155)= 192.7, p<.001; model correctly classified 98.7% of cases. At step 2, covariates predicted an additional 0.9% of its variance increasing it to 72.1%, χ²(4) (N=155)= 197.9, p<.001; model correctly classified 98.1% of cases. At step 3 (χ²(6) (N=155)=209.4, p<.001), IVs predicted an additional 2% of its variance increasing it to 74.1% ; model correctly classified all cases, see Table 7. None of the IVs predicted T2 obesity category after controlling T1 obesity category, demographics and covariates.
Mediational analyses examined whether impaired sleep mediated the relationships between the measures of behaviour and body fatness. Of all the potential IVs, only NEQ-nocturnal ingestions (NI) met criterion as an IV and only three sleep symptoms (i.e. SSQ, SD, DD) met criteria as mediators. That is, SSQ, SD and DD were correlated with NEQ-NI and in turn they were correlated with at least one body fatness measure at T1, see correlations in the supplementary file. Thus, only they were tested as potential mediators. Results showed that high NI was related to high SD, B = .07, SE = .02, 95% CI [.03,.10], β = .28, p < .001, and high SD predicted high BMI, B = 2.11, SE = 1.02, 95%CI[.11,4.12], β = .17, p = .04, supporting Hypothesis 2. NI no longer predicted BMI after controlling for the effects of the mediator, SD, B = .10, SE = .24, 95%CI[-.39,.58], β = .03, p = .70, thus, SD fully mediated the NI-BMI relationship accounting for about 3.3% of the variance in BMI (R2 = .033). Indirect effects were tested using a percentile bootstrap estimation approach with 5,000 samples and the PROCESS macro-Version 4 (Hayes, 2022). Results showed that the indirect coefficient was significant, B = .14, SE = .08, 95% CI [.02,.32], standardized β = .05 suggesting that more NI may have contributed to high BMI via more SD in some participants.
Mediational analysis examined if SSQ mediated the NI - WHR relationship. Results showed that high NI was related to poor SSQ, B = .08, SE = .03, 95%CI[.03,.13], β = .24, p = .002, and SSQ predicted high WHR, B = -.02, SE = .01, 95%CI[-.04,-.002], β = -.17, p = .03, supporting Hypothesis 2. NI no longer predicted WHR after entering the mediator, SSQ, B = -.002, SE = -.003, 95%CI[-.008,.004], β = -.05, p = .50, suggesting that poor SSQ fully mediated the NI – WHR relationship, predicting about 1% of the variance in BMI (R2 = .01). Indirect effects were tested using a percentile bootstrap estimation approach with 5,000 samples and the PROCESS macro-Version 4 (Hayes, 2022). Results showed that the indirect coefficient was significant, B = -.002, SE = .001, 95%CI [-.003, -.0001], standardized β = .04, suggesting that more NI may have contributed to greater WHR via a drop in SSQ.
Mediational analysis examined if DD mediated the NI - obesity category relationship. Results showed that high NI was related to high DD, B = .07, SE = .03, 95%CI[.02,.12], p = .01, and DD predicted obesity category, B = -.63, SE = .23, 95%CI[-1.08,-.17], p = .007, supporting Hypothesis 2. NI no longer predicted obesity category after entering the mediator DD, B = .14, SE = .08, 95%CI[-.002,.30], p = .053, suggesting that high DD fully mediated the high NI – obesity category relationship, predicting about 5.9% of its variance (CoxSnell R2 = .059). Indirect effects were tested using a percentile bootstrap estimation approach with 5,000 samples and the PROCESS macro-Version 4 (Hayes, 2022). Results showed that the indirect coefficient was significant, B = -.04, SE = .03, 95%CI [-.1, -.01] suggesting that more NI may have contributed to an increase in DD which in turn may have contributed to more participants being categorised as overweight/obese.