Body weight

weight BODY_WEIGHT_0

Descriptive statistics (Continuous)

Provides descriptive statistics for a subset or all continuous variables in the study data

Data type mismatch

Check if the declared data types in the metadata match the observed data types in the study data.

Presence and amount of missing data for each variable, separated by class of missingness (i.e., system missingness (NA), used missing codes and jump codes). The higher the number or percentage of missing values, the lower the data quality.
Note:
1. “ADDED: SysMiss” or “Sysmiss N” refers to NA’s (i.e., system-indicated missing values);
2. The percentage of all the columns in the table (except Measurements N (%)) are calculated over the Observation N. The percentage of Measurement N is calculated by dividing the value (Measurement N) by the difference between the number of observations and the number of jumps (Observation N - Jumps N).

Response-rates (Item-level)

The higher the rate, the lower the data quality. For further information, see here

Range violations

Check applied to numerical or time-date variables. If a range of values is provided in the metadata, the presence of values outside the interval is checked. These can be: inadmissible values (hard limits), improbable but plausible values (soft limits), or values outside measurement ranges (detection limits). The higher the number or percentage of values outside the limits, the lower the data quality.
Attention: values outside hard limits are removed from the following quality checks.

Cumulative distribution device

stratified ECDF Plots help find observer differences

Cumulative distribution observer

stratified ECDF Plots help find observer differences

Distribution

Distribution plots for the variables in the study data (distributions are described using bar plots in case of categorical data and histograms for numerical data).

End-digit preference

Checks if the distribution of end-digits of a numerical variable follows a uniform distribution, where each digit has a probability of 0.1.
Recommended for manually recorded values, because humans tend to round values.

Time trend device

Lowess regression models are used to display time trends and potential effects of instruments used or examiners on the measurements. Background knowledge is needed to interpret the graph. Typically the further the graphs for different devices or examiners are apart from each other, the bigger the instruments or examiners effect.

Time trend observer

Distribution across device

Marginal distribution per device or observer and overall distribution are reported to check for the presence of an effect of the instrument or observer on the measurements obtained. Background knowledge is needed to interpret the graphs.

Distribution across observer

Univariate outliers

Presence of outliers in the study data are checked using 4 approaches (Tukey, 3SD, Hubert, and SigmaGap). For each value it is indicated how many approaches identifies it as outlier (0 = not an outlier, 4 = an outlier based on 4 approaches).
The higher the number or percentage of identified outliers, the lower potentially the data quality.

Variance proportion observer

Checks the effect of different devices or observers on measurements through variance based models and intra-class correlations (ICC).
The closer to 1 the ICC value is, the lower the data quality.
Note: this check employs a model.