Participants
In total, 160 parents filled out baseline questionnaires (test) and 100 parents filled out the follow-up questionnaires (retest). The average number of days between test and retest was 64.6 days (sd ± 39.2 days). Seventy percent of the questionnaires were answered by mothers. Mean age for the children was 6.9 years old (sd ± 2.2 years old, range 3.0–10.7 years old). Forty-four percent of the children were girls.
Exploratory and confirmatory factor analyses
After a series of exploratory factor analyses, we found that the communality (common variance with other variables) of item 6 (hours of night sleep) was low, and accordingly, it was excluded in further analyses. The final exploratory model was found to have two factors: sleep onset problems (SOP) (item 1 “My child takes longer than 30 minutes to fall asleep after going to bed” and item 2 “My child has problem falling asleep at bedtime”) and sleep maintenance problems (SMP) (item 3 “My child awakes more than once during the night”, item 4 “After waking during the night, my child has trouble returning to sleep” and item 5 “My child appears sleepy during the day”). From confirmative factor analyses, which were based on the exploratory factor model, we found that two dimensions are needed to account for the common variance between the five variables of the PISI.
Figure 1A and 1B present the confirmatory analysis two-factor solutions, SOP and SMP, for baseline and test-retest, respectively. Both models showed a good fit. The fit was χ2 = 0.43, df = 3, p = 0.93, RMSEA = 0.00, and CFI = 0.99 at baseline, and the fit was χ2 = 0.23, df = 2, p = 0.89, RMSEA = 0.00, and CFI = 0.99 at test-retest. As can be seen, SOP and SMP are positively correlated (baseline r = 0.27, and test-retest r = 0.38). The construct reliability for SOP and SMP at baseline was 0.86 and 0.62, respectively. The corresponding value for SOP and SMP at test-retest was 0.71 and 0.76, respectively, indicating that the construct reliability of the Swedish version of the PISI is reliable and replicable.
To further analyse the construct validity and reliability, we explored (using SEM) how the SOP and SMP at baseline was associated with SOP and SMP at retest.
Figure 2 shows that the model has a good fit (i.e., χ2 = 30.20, df = 24, p = 0.18, RMSEA = 0.05, and CFI = 0.98), and SOP and SMP at baseline were highly correlated with SOP and SMP at test-retest (r = 0.71 and r = 0.72, respectively). Thus, SOP and SMP at baseline have a substantial effect or predictive power on SOP and SMP at test-retest. More than 50% of the true variance in SOP and SMP at test-retest can be explained by the variance of the factors at baseline. The baseline/test-retest correlations also support the reliability of the factors in the PISI.
To make the factors practicable, the means of the variables of each factor in the PISI have been calculated (with equal weight of the variables) and then correlated. As can be seen, the correlations of Figure 3 are in correspondence (r = 0.66 and r = 0.77, respectively) with the model in Figure 2 (r = 0.71 and r = 0.72, respectively).
It is possible that the child’s age may influence the parent’s response in the PISI. Therefore, we performed a re-analysis on the model in Figure 2 and controlled for age by means of partial correlation analysis. The results showed that the model was stable, thus the age of the children had no influence on the model.
Taken all together, this indicates that the two-dimensional structure of the Swedish version of the PISI has substantial construct and test-retest reliability.
Criterion validity of the PISI and KIDSCREEN-27
To explore the criterion validity of the PISI, we analysed the correlations between the two factors in the PISI (SOP and SMP) from the baseline measurement and test-retest measurements to the five dimensions in KIDSCREEN-27. The correlations were optimized by means of confirmative factor analyses. The correlations between SOP and SMP from the two data collection points and the five dimensions in KIDSCREEN-27 were generally weak and non-significant for SOP (Table 1). But, SMP, on the other hand, correlated significantly with all dimensions in KIDSCREEN-27. However, SOP and SMP are correlated, and it can be reasonable to assume that the former affects the latter. Thus, problems with falling asleep in the evening (i.e., SOP) may cause sleeping problems during the night (i.e., SMP) (Figure 1).
Table 1. Optimally weighted correlationsa between SOP and SMP and the five criterion dimensions of KIDSCREEN-27.
|
|
School environment
|
Psychological
well-being
|
Autonomy and parent relations
|
Social support
and peers
|
Physical
well-being
|
|
|
r
|
p-value
|
r
|
p-value
|
r
|
p-value
|
r
|
p-value
|
r
|
p-value
|
Sleep
Onset
|
Test
|
-.07
|
.38
|
-.17*
|
.03
|
-.14
|
.09
|
-.16*
|
.04
|
-.06
|
.46
|
Problems
|
ReTest
|
-.10
|
.35
|
-.17
|
.10
|
-.04
|
.71
|
-.02
|
.85
|
.13
|
.22
|
Sleep Maintenance
|
Test
|
-.40*
|
<.01
|
-.36*
|
<.01
|
-.10
|
.23
|
-.23*
|
.04
|
-.31*
|
<.01
|
Problems
|
ReTest
|
-.47*
|
<.01
|
-.48*
|
< .01
|
.52*
|
<.01
|
-.26*
|
.01
|
-.42*
|
<.01
|
a Pearson correlation coefficient (r)
* Significant correlations (p <.05)
Therefore, we performed a series of SEM analyses using SOP and SMP. In these models, we assumed that SOP had an effect on SMP, which in turn, had effects on HRQoL (i.e., the different dimensions in KIDSCREEN-27).
Figure 4 shows the model for PWB. The model had a good fit (χ2 = 51.83, df = 41, p = 0.19, RMSEA = 0.04, and CFI = 0.97) and showed that SMP has a direct effect (B = -0.49), indicating a decreasing effect on PWB. For SOP, there was a direct effect (B = 0.52) on SMP, indicating that SOP increases SMP, and also an indirect negative effect on PWB (B = -0.26), indicating that SMP is a mediating factor between SOP and PWB.
Table 2 presents the indirect and direct effects from SOP and SMP on the dimensions of KIDSCREEN-27. As can be seen, SOP and SMP had effects on all dimensions of the KIDSCREEN-27. The models showed that there were significant direct effects of SMP on the criterion measures and significant indirect effects of SOP on the criterion measures. However, in the SOC dimension, no significant indirect effect of SOP could be found. The predictive power (i.e., the ability to “explain” the variance of the criterion dimensions) of the two factors ranged from 7% to 27% (23% for PSY, 18% for PHY, 27% for PAR, 7% for SOC, and 22% for SCH).
When scrutinizing the KIDSCREEN-27-dimensions, we found a mean correlation between the five dimensions of .44, and accordingly, a “second order factor” was to be expected. In a confirmative second order factor analysis, we found that the five dimensions form a second order general KIDSCREEN-27 factor, and SMP (test-retest) has a factor loading of -.48 with the general KIDSCREEN-27 factor (χ2 = 8.42, df = 9, p = 0.49, RMSEA = 0.00, CFI = 0.99). Thus, the SMP dimension is directly or indirectly related to all five KIDSCREEN-27 factors and explains 23% of the variance of the general KIDSCREEN-27 factor.
Table 2. Correlations between the PISI and KIDSCREEN-27.
Criterion-dimension “Re-test”
|
Physiological well-being (PHY)
|
Autonomy and parent relations (PAR)
|
Social support and peers (SOC)
|
School - environment (SCH)
|
Psychological well-being (PWB)
|
Effects “Re-Test”
|
Indirect
Effect
|
Direct
Effect
|
Indirect
Effect
|
Direct
Effect
|
Indirect
Effect
|
Direct
Effect
|
Indirect
Effect
|
Direct
Effect
|
Indirect
Effect
|
Direct
Effect
|
SOP “Re-Test”
|
-.14
|
|
-.26
|
|
-.19 (n.s.)
|
|
-.21
|
|
-.26
|
|
SMP “Re-Test”
|
|
-.32
|
|
-.47
|
|
-.19
|
|
-.45
|
|
-.49
|
Model fit- indices
Chi-square
df
P
RMSA
CFI
|
|
|
|
|
|
38.06
|
56.45
|
29.00
|
24.21
|
51.83
|
30
|
44
|
25
|
22
|
41
|
.15
|
.10
|
.25
|
.37
|
.19
|
.05
|
.05
|
.04
|
.03
|
.04
|
.97
|
.97
|
.98
|
.98
|
.97
|
Direct and indirect effects from structural equation models of the factors sleep onset problems (SOP), and sleep maintenance problems (SMP) on the KIDSCREEN-27 domains Physiological well-being, Autonomy and parent relations, Social support and peers, School environment and Psychological wellbeing. The figures in the table are based on the models from the re-test-situation. n.s = non-significant