2024-10-07
The MoBa reproducibility project assesses current open science practices in the Norwegian Mother, Father, Child cohort (MoBa) and provides guidelines on being as transparent science practices given the limitations of epidemiological research.
This MoBaRepro repository is a supplement to the working example provided in the paper.
In this hypothetical research project, researchers are interested in testing the independent effects of age and breastfeeding duration on height during childhood in MoBa. Additionally, they also intend to test for variation in these effects according to child sex.
H1: Higher age predicts taller height across childhood. H2: Longer breastfeeding duration will be associated with taller height across childhood. H3: The effect of age on height across childhood differs between males and females. H4: The effect of breastfeeding duration on height across childhood differs between males and females.
The data included in this repository is completely simulated, and bears no resemblance to actual MoBa data other than in structure.
All scripts can be found under the scripts subdirectory.
data_sim.R generates the simulated “original” MoBa dataset with
variables for height at 6 months, 18 months, 3 years, 5 years, 7 years,
and 8 years, as well as for breastfeeding duration, sex, and birth
length. This dataset is subsequently saved in the data subdirectory as
simdata.RData.
The 01_dataprep.R script loads the simulated dataset, checks the
descriptive statistics and distributions, cleans and recodes the data,
and converts the data to long form before the analysis. The processed
dataset is saved as data/longdata.RData.
The 02_mixedmodels.R script loads the longdata.RData dataset
generated in script 01, which is used to create a single multilevel
model to test all hypotheses. It also performs two sensitivity analyses:
one using a different method of categorising the breastfeed_dur
variable, and one using the height variable where values were deemed
outliers if they were more than 2 SD away from the mean, and were thus
removed. Finally, this script collates the results from all models.
Because MoBa data contains potentially identifiable information, it is
not possible to share the actual data alongside analysis scripts. One
way of handling this issue is to create a synthetic dataset based on the
original dataset. The 03_synthesis.R script provides an example of how
to generate such a synthetic dataset and how to assess its preservation
of the relationships found in the original dataset. It loads the
longdata.RData dataset and creates a synthetic dataset,
data/syndata.RData.
The 04_validate.R script performs the same analyses as in script 02,
but on the synthetic dataset.