Skip to contents

7. Describing the Data Without Knowing Sleep

Source: vignettes/chapter7_DescribingDataWithoutKnowingSleep.Rmd
chapter7_DescribingDataWithoutKnowingSleep.Rmd

With the metric values imputed in the previous chapter, GGIR part 2 offers a first descriptive analysis of the data. Although we will have to wait till later GGIR parts (chapters) to see the segmentation of days in waking and sleep hours, there is already enough that can be quantified at this point. In this chapter we focus on descriptives of the data that are informative even without the knowledge on when the participant was sleeping or awake.

Data quality indicators

GGIR part 2 summarises all the data quality checks done in the previous four chapters, ranging from a report on the successfulness of the auto-calibration procedure to the number of valid days. In this way, GGIR part 2 is an ideal place to start for data quality assurance.

Basic descriptives

Descriptive variables are calculated and reported for valid days only, where the criteria for valid day is defined by parameter includedaycrit.

Average acceleration

Average acceleration is known to be correlated with the activity-related energy expenditure. GGIR part 2 provide two types of average acceleration:

  • Average per day, only stored when the day was considered valid. Note that this descriptive and most other descriptives are also stored by GGIR as averages across all days, weekend days or weekday, which we will discuss in more detail.

  • Weighted average of all valid data points in the recording, weighted by timing in the day of all valid epochs, regardless of whether they come from days that as a whole are classified as valid or not.

Acceleration distribution

To get a more detailed description of the data looking at acceleration value distribution can be informative, which we GGIR facilitates in two ways :

  1. By specifying the quantiles of the distribution with parameter qlevels, which are fed to build-in R function quantile and give us the metric values corresponding to such quantiles (a quantile multiple by 100 is the same as a percentile).

  2. By describing time spent in acceleration ranges, which are defined by parameter ilevels .

In the physical activity literature the distribution of acceleration values is often referred to as intensity distribution.

Derived descriptives

Sets of quantiles (MX metrics by Rowlands et al.)

The quantiles as discussed above can be used to describe the accelerations that participants exceed in their most active “X” accumulated minutes in a day. In the specific approach as proposed by Rowlands et al. these quantiles are referred to as the MX metrics. The MX metrics should not be confused with the most active continuous X hours, e.g. M10, as used in circadian rhythm research that also can be derived with GGIR (see parameter winhr).

To use the MX metrics as proposed by Rowlands et al., specify the durations of the 24h day that you wish to filter out the accelerations for. For example, to generate the minimum acceleration value for the most active 30 minutes, you can call qlevels = (1410/1440), which will filter out the lowest 1410 minutes of the day. This can also be used as a nested term to generate multiple metrics, for example to call M60, M30 and M10, you can use the following parameter:

qlevels = c(c(1380/1440), c(1410/1440), c(1430/1440)).

Note: if parameter qwindow has values less than 24 h, e.g. the school day (e.g. Fairclough et al 2020), the denominator should be changed from 1440 (24h) to the new qwindow window length, e.g. if an 8-hour window is used then the denominator would be 480 rather than 1440 and the numerator would be 480 minus for example 60, 30, 10 or 5.

The output in the part 2 summary files will refer to this as a percentile of the day. Thus, for a 24-h day, M30 will appear as “p97.91666_ENMO_mg_0.24hr”. To create the radar plots of these MX metrics as first described by Rowlands et al., this GitHub repository provides the R code and detailed instructions on how to make the radar plots using your own data.

Intensity gradient

If we plot the time spent in equally spaced acceleration ranges we would end up with an asymptotic-shaped curve, indicating little time spent at high intensities (acceleration levels) and much time spent at low intensities. The shape of the distribution may be informative but is hard to quantify with single number in its standard form. Therefore, a new concept called the intensity gradient was proposed by Rowlands and colleagues.

The intensity gradient defines the slope of the log-transformed axes in the distribution. More specifically, we calculate the time accumulated in incremental acceleration bins (bin size = 25 mg) but also keep track of the mid-point of each intensity bin, e.g. 62.5 mg for the bin ranging from 50 to 75 mg. Both the mid-point acceleration of a bin expressed in mg and the time spent in a bin expressed in minutes are then log-transformed. The log-transformation is expected to change the asymptotic-shaped curve into a straight line. Subsequently, a linear regression is fitted through these data points. The slope of this regression line represents the intensity gradient. Further, we calculate the correlation coefficient for the data points to help verify the degree to which they form a straight line.

The intensity gradient is not calculated by default. To include this metric in the part 2 output, set iglevels = TRUE, which will ensure that the algorithm is used.

Further, if you want to do more methodological research on this, you can use this parameter to define alternative acceleration bins , e.g. for using bins of 20 instead of 25 mg iglevels = c(seq(0, 4000, by = 20), 8000).

Key arguments

  • includedaycrit

  • ilevels

  • qlevels

  • iglevels

  • qwindow

  • do.report

GGIR part 2 generates three csv reports: part2_daysummary.csv, part2_summary.csv, and data_quality_report.csv. As data_quality_report.csv was discussed in chapter 3 we will focus here only on the first two reports.

The variables in part2_summary.csv are the recording level aggregates of the variables in part2_daysummary.csv. Here, variable names starting with the “AD_” refer to average across all days, “WD” refers to the average across weekdays, “WE” refers to the average across weekend days, “WWE” refers to weighted weekend days to ensure both weekend days contribute equally, “WWD” refers to weighted weekdays to ensure all weekdays contribute equally.

Further, GGIR part 2 generates a report named part2_daysummary_longformat.csv, which is generated when GGIR is used for day segment analysis, see documentation for parameter qwindow. This report contains the exact same information as part2_daysummary.csv, yet in long format. In part2_daysummary_longformat.csv, each row represents one segment from one day in one recording, while in part2_daysummary.csv, each row contains one day in one recording and the segments of the day are organized in different columns.

Descriptive variables

To clarify that b refers to both part2_summary.csv and part2_daysummary.csv, r refers to part2_summary.csv only.
(Part of) variable name Description Report(s)
mean_ENMO_mg_0-24hr average acceleration defined by metric ENMO for the full day b
ENMO_fullRecordingMean Weighted average over acceleration across recording r
p20_ENMO_mg_0-24h 20th percentile based on the full day b
p20_ENMO_mg_0-24h_fullRecording 20th percentile based on the weighted average day from the full recording r
[0,50)_ENMO_mg_0-24hr time spent in minutes between 0 and 50 m_g_ acceleration defined by metric ENMO for the full day. Square bracket means that the value itself is included, round bracket indicated that value itself is not included b
ig_gradient_ENMO_0-24hr Gradient for the intensity gradient b
ig_intercept_ENMO_0-24hr Intercept for the intensity gradient b
ig_rsquared_ENMO_0-24hr R-squared for the intensity gradient b