Fig 1.
The relationship between daily sleep log and wrist actigraphy measurements.
Bivariate histograms for (a) sleep onset (SON), (b) sleep offset (SOFF), and (c) total sleep time (TST). The white dotted line represents the equality line (perfect agreement); therefore, deviations from the line indicate a lack of agreement. The empirical distributions (right panels) show the distribution of differences (actigraphy minus sleep log values), with the black dotted line representing the bias, and boxes demarcating the three reference intervals included in our non-parametric Bland-Altman analysis (±0.5, ±1.0, ±1.5 hours).
Table 1.
Group mean results for six variables measured by wrist actigraphy and sleep logs.
Fig 2.
Example data comparing daily sleep onset times from actigraphy and sleep logs.
The time series’ in (a) show a participant with strong agreement, and (b) show a participant with relatively weak agreement. Green bars represent the difference between the measures (actigraphy–sleep log). The right panels illustrate difference distributions across all measurements for that participant.
Table 2.
Individual differences in bias, agreement, compliance and demographic variables.
Fig 3.
Compliance across time and in relation to agreement and personal traits.
(a) Group compliance is shown across the 16-week data collection. The proportion of participants who completed the daily questionnaire is shown in red, and the amount of time delay between self-reported sleep offset and sleep log submission time is shown in blue. Solid lines represent best-fit linear models to the data. (b) Scatter plots illustrate the relationship between an individual’s mean compliance rate (left) and time delay (right) and their level of agreement (SON and SOFF combined). Agreement was defined as the percentage of data with an absolute difference less than one hour. (c) Scatter plots illustrate compliance rate and time delay in relation to behavioral avoidance measures derived from BIS/BAS scale.
Table 3.
Pearson’s correlation coefficients for compliance variables and personality traits.
Fig 4.
Time series for an example participant showing actigraphy, sleep logs, and model output.
Green dots represent the combined estimate for sleep onset time, which was computed as a weighted sum of the actigraphy and sleep log measurements. The text with arrows highlight days exemplifying good agreement, poor agreement, and non-compliance due to a missing sleep log. The upper violin-style plots represent the empirical distribution of all sleep onset measurements from actigraphy and sleep logs. The model was designed for robustness in cases where actigraphy and sleep logs have discrepant measurements (upper left) by producing a composite estimate closer to the more likely value considering the individual’s sleep history. When measurements have similar probability (upper right), the model produces a value that is near the average of the two measurements.
Fig 5.
Narrowing of difference distribution and outlier reduction for combined estimates.
Bivariate scatter plots for (a) sleep onset time, and (b) sleep offset time, showing actigraphy measurements (red) and combined estimates (green) with reference to sorted daily sleep log measurements (blue) across the entire data set. The difference distributions (right) are much sharper for combined estimates, with corresponding increase in agreement (e.g. a higher percentage of differences contained within ±1.0 hour).