identify_gaps {mpathsenser}

Identify gaps in mpathsenser mobile sensing data

Description

Oftentimes in mobile sensing, gaps appear in the data as a result of the participant accidentally closing the app or the operating system killing the app to save power. This can lead to issues later on during data analysis when it becomes unclear whether there are no measurements because no events occurred or because the app quit in that period. For example, if no screen on/off event occur in a 6-hour period, it can either mean the participant did not turn on their phone in that period or that the app simply quit and potential events were missed. In the latter case, the 6-hour missing period has to be compensated by either removing this interval altogether or by subtracting the gap from the interval itself (see examples).

Usage

identify_gaps(
  db,
  participant_id = NULL,
  min_gap = 60,
  sensor = "Accelerometer"
)

Arguments

db

A database connection to an m-Path Sense database.

participant_id

A character string identifying a single participant. Use get_participants to retrieve all participants from the database. Leave empty to get data for all participants.

min_gap

The minimum time (in seconds) passed between two subsequent measurements for it to be considered a gap.

sensor

One or multiple sensors. See sensors for a list of available sensors.

Details

While any sensor can be used for identifying gaps, it is best to choose a sensor with a very high, near-continuous sample rate such as the accelerometer or gyroscope. This function then creates time between two subsequent measurements and returns the period in which this time was larger than min_gap.

Note that the from and to columns in the output are character vectors in UTC time.

Value

A tibble containing the time period of the gaps. The structure of this tibble is as follows:

participant_id	the participant_id of where the gap occurred
from	the time of the last measurement before the gap
to	the time of the first measurement after the gap
gap	the time passed between from and to, in seconds

Warning

Depending on the sensor that is used to identify the gaps (though this is typically the highest frequency sensor, such as the accelerometer or gyroscope), there may be a small delay between the start of the gap and the actual start of the gap. For example, if the accelerometer samples every 5 seconds, it may be after 4.99 seconds after the last accelerometer measurement (so just before the next measurement), the app was killed. However, within that time other measurements may still have taken place, thereby technically occurring "within" the gap. This is especially important if you want to use these gaps in add_gaps since this issue may lead to erroneous results.

An easy way to solve this problem is by taking into account all the sensors of interest when identifying the gaps, thereby ensuring there are no measurements of these sensors within the gap. One way to account for this is to (as in this example) search for gaps 5 seconds longer than you want and then afterwards increasing the start time of the gaps by 5 seconds.

Examples

## Not run: 
# Find the gaps for a participant and convert to datetime
gaps <- identify_gaps(db, "12345", min_gap = 60) |>
  mutate(across(c(to, from), ymd_hms)) |>
  mutate(across(c(to, from), with_tz, "Europe/Brussels"))

# Get some sensor data and calculate a statistic, e.g. the time spent walking
# You can also do this with larger intervals, e.g. the time spent walking per hour
walking_time <- get_data(db, "Activity", "12345") |>
  collect() |>
  mutate(datetime = ymd_hms(paste(date, time))) |>
  mutate(datetime = with_tz(datetime, "Europe/Brussels")) |>
  arrange(datetime) |>
  mutate(prev_time = lag(datetime)) |>
  mutate(duration = datetime - prev_time) |>
  filter(type == "WALKING")

# Find out if a gap occurs in the time intervals
walking_time |>
  rowwise() |>
  mutate(gap = any(gaps$from >= prev_time & gaps$to <= datetime))

## End(Not run)