Compare within-person and between-person correlations
Source:R/stats_correlate_levels.R
correlate_levels.RdComputes two distributions of correlations for every pair of variables, following Fisher, Medaglia and Jeronimus (2018):
Arguments
- data
Long-format data frame.
- id
Person identifier column.
- vars
Variables to correlate (at least two).
- time
Optional column giving the order of occasions within persons.
- lag_adjust
Remove first-order temporal dependence before computing within-person correlations.
- min_n
Minimum number of complete pairs for a correlation to be computed, within a person or across persons at one occasion.
Value
A data frame of class idiographic_level_correlations, one row per
correlation, with columns level ("within" or "between"), unit
(the person for "within", the occasion number for "between"), x,
y, n (complete pairs), and r. Summaries per pair are returned by
summary() and by as.data.frame(x, what = "summary").
Details
within persons (intraindividual): one correlation per person, computed across that person's occasions;
between persons (interindividual): one correlation per occasion, computed across persons. Each person's series is ordered by
time; the first observations of all persons form the first cross-sectional data set, the second observations form the second, and so on.
If group-level results generalise to individuals, the two distributions have similar means and similar spreads. A within-person spread that is several times larger than the between-person spread indicates that the group correlation describes few individuals well.
With lag_adjust = TRUE, within-person correlations are computed on
residuals after regressing each variable on its own lag-1 value and the
lag-1 value of the other variable, which removes first-order temporal
dependence. The original authors repeated this step until the Durbin-Watson
test was non-significant; a single first-order adjustment is used here.
Between-person correlations are always computed from the observed values.
References
Fisher, A. J., Medaglia, J. D., & Jeronimus, B. F. (2018). Lack of group-to-individual generalizability is a threat to human subjects research. Proceedings of the National Academy of Sciences, 115(27), E6106-E6115. doi:10.1073/pnas.1711978115
Examples
lv <- correlate_levels(srl, id = "name", vars = c("efficacy", "effort"),
time = "day")
summary(lv)
#> x y n_within mean_within median_within sd_within n_between
#> 1 efficacy effort 36 0.3637722 0.3587067 0.2498384 156
#> mean_between median_between sd_between mean_difference sd_ratio
#> 1 0.5034717 0.5096236 0.1373181 -0.1396996 1.819413