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.
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.
Descriptive variables are calculated and reported for valid days
only, where the criteria for valid day is defined by parameter
includedaycrit
.
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 weekdays, which we will discuss in more detail later on.
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.
To get a more detailed description of the data, looking at the distribution of acceleration values can be informative. GGIR facilitates this in two ways:
By specifying the quantiles of the distribution with parameter
qlevels
, which are fed to the build-in R function
quantile
, GGIR gives us the metric values corresponding to
such quantiles (a quantile multiplied by 100 is the same as a
percentile).
By describing the time spent in acceleration ranges, which are
defined by parameter ilevels
.
The distribution of acceleration values is often referred to as intensity distribution in the physical activity literature.
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 during which you want to identify the
accelerations values. For example, to generate the minimum acceleration
value for the most active accumulated 30 minutes, you can call
qlevels = (1410/1440)
. This parameter also accepts nested
terms to generate multiple MX metrics. For example, to call M60, M30,
and M10, you can specify the following:
qlevels = c(c(1380/1440), c(1410/1440), c(1430/1440))
.
Note: If time segments shorter than 24 hours are specified in
parameter qwindow
, such as the 8-hour school day (as
described in Fairclough et al
2020), the denominator in qlevels
should change from
1440 (24h) to the specific segment length. In this example, we would use
480 (8h). Accordingly, the argument to call M60, M30, and M10 would
be:
qlevels = c(c(1380/1440), c(1410/1440), c(1430/1440))
.
At the moment, this only works for one segment length and GGIR does not facilitate the generation of MX metrics for multiple unequal time segments within the same GGIR function call.
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.
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 a 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 of this intensity 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 (R^2).
The intensity gradient is not calculated by default. To include this
metric in the part 2 output, set iglevels = TRUE
.
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)
.
includedaycrit
ilevels
qlevels
iglevels
qwindow
do.report