Statistical analysis

Participants were included in the analyses only if they had usable data for the baseline measurement and at least one data collection week during the lighting intervention. Two participants were excluded from the actigraphy analysis because they did not wear the actigraph during one or more data collection weeks. Fifteen participants did not complete all 25 weeks of the study (Supplementary Table 2).

Linear mixed-effects models (LMEMs) were constructed for data analysis using SPSS, version 24.0, for Windows (SPSS Inc., Chicago, IL). The criterion for statistical significance in all analyses was p < 0.05. While the central focus of the study’s primary and secondary aims was the efficacy of the TLI, we also assessed the statistical effects of participants’ sex and level of dementia (cognitive state). For the cognitive state variable, participants were placed into one of two categories (i.e., mild–moderate dementia or severe dementia) based on their BIMS or MMSE scores, when available (see Participant Selection). (The mild and moderate categories were merged due to the small number of participants in the moderate range [n = 3]). No BIMS or MMSE scores were available for nine participants. However, as these individuals resided in an assisted living facility where the admission criteria were limited to residents with mild–moderate dementia, explicitly excluding those determined to have severe dementia, the nine participants without BIMS or MMSE assessments were coded as mild–moderate for purposes of the data analysis.

Although initially considered for inclusion in the analysis, participant age was ultimately excluded as a variable because Pearson’s product-moment correlations between age and the outcomes revealed nonsignificant results (p > 0.05) for all but one of the measures (CSDD at week 3 only).

Data collection week (i.e., at baseline and during the lighting intervention), sex, and cognitive state were entered into the model as fixed effects and participant was entered as a random effect. Significant effects due to main factors (i.e., data collection week, sex, and cognitive state) and/or their interactions were followed up with multiple comparisons using two-sided t-tests with Sidak correction for Type I errors. Cohen’s d is reported as a measure of effect size for significant differences. Estimated mean and standard error of the mean (SEM) values are reported.