Individual-level data from a Solomon six-group study of interventions to support the transfer of time-management training: pretesting crossed with relapse prevention, proximal plus distal goal setting, and a control condition (Mai et al., 2020). The data were published as the article's supplementary material and are redistributed under its Creative Commons Attribution 4.0 license (see License).
Format
A data frame with 211 rows, one per participant, and 6 variables:
- id
Row number in the published file.
- gender
Female or Male.
- pretested
1 if assigned to be pretested, 0 if not.
- condition
RP(relapse prevention),GS(proximal plus distal goal setting), orControl.- pre_behavior
Self-reported time-management behavior before the workshop (pretested groups only).
- post_behavior
Self-reported time-management behavior after the workshop (see Details for the timing); missing for participants who did not answer.
Source
Mai, N. N., Takahashi, Y., & Oo, M. M. (2020). Testing the effectiveness of transfer interventions using Solomon four-group designs. Education Sciences, 10(4), Article 92. https://doi.org/10.3390/educsci10040092 (supplementary material, Table S1: dataset.sav)
Details
Design. Final-year students at a management college took a three-hour time-management workshop.
Pretesting. Before the workshop, they were "randomly assigned into either the pretested or unpretested groups based on their roll numbers" (p. 5). The article does not say how roll numbers were used.
Interventions. After it, they were randomly assigned to relapse prevention (RP), goal setting (GS), or control (p. 6).
Measures. Self-reported time-management behavior was measured before the workshop in the pretested groups and 10 weeks after it in all groups (p. 7; the note to Table 3 on p. 8 says eight weeks).
Every pair of conditions forms a Solomon four-group design. The authors
analyzed three such pairs, reusing groups across them. fit_solomon_glm()
with control = "Control" fits one model to all six groups (see the
examples and the article "Designs With Several Treatments").
Sample and attrition. The file has 211 participants, and the article reports 210 (p. 5). 133 answered the posttest, with attrition that differed by group (see the example). The unpretested groups have no pretest by design.
Scale. pre_behavior and post_behavior are the authors' scale
scores for the 30 self-report items. The article describes the response
scale as running from 1 (always) to 5 (never) (p. 7). The value labels in
the data file run from 1 (never) to 5 (always). The scores are not the
simple means of the 30 items in the file, so the items are not included.
Known results. The two-by-two ANOVAs of the posttest scores reproduce the authors' Table 4 (p. 8). For relapse prevention against control, the Pretest x Treatment sum of squares is 0.508, the error sum of squares 13.347, and F = 3.461. The history checks of their Table 5 (p. 8) and the ANCOVAs of the first two comparisons in their Table 7 (p. 9) are reproduced as well. The third ANCOVA, relapse prevention against goal setting, gives the published difference of -0.219 with p = .044 against the published .046. The published cell statistics of their Table 6 do not match the data (see issue #53). The article "Worked Example: A Published Solomon Study" carries the relapse-prevention comparison through the package's analyses.
Acknowledgment
We thank Nu Nu Mai, Yoshi Takahashi, and Mon Mon Oo for making their data publicly available with their article. For the study's design, measures, and findings, read the published article (Mai et al., 2020, cited under Source).
License
The data are © 2020 by the authors (Nu Nu Mai, Yoshi Takahashi, and Mon Mon
Oo) and are published with the
article under the Creative Commons Attribution 4.0 International license
(https://creativecommons.org/licenses/by/4.0/). MDPI's open access
policy (https://www.mdpi.com/openaccess) states that its articles,
"including their data, graphics, and
supplementary material", may be reused with attribution. Changes made in
solomonR: 6 of the 117 variables in the published file are kept, renamed,
and recoded to factors or 0/1 indicators, and an id column is added. The
original file and the script that builds this data set are in the
package's data-raw folder on GitHub. See also inst/COPYRIGHTS.
Examples
# Posttest completion by group.
with(mai2020, table(pretested, condition, posttest = !is.na(post_behavior)))
#> , , posttest = FALSE
#>
#> condition
#> pretested RP GS Control
#> 0 9 12 13
#> 1 11 10 23
#>
#> , , posttest = TRUE
#>
#> condition
#> pretested RP GS Control
#> 0 22 15 22
#> 1 24 23 27
#>
# Relapse prevention against control, as a Solomon four-group design.
rp <- subset(mai2020, condition %in% c("RP", "Control"))
with(rp, fit_solomon_glm(post_behavior, treat = as.integer(condition == "RP"),
pretested = pretested, pretest_score = pre_behavior))
#> Warning: The `pretest_score` argument of `fit_solomon_glm()` is deprecated as of
#> solomonR 0.8.0.
#> ℹ Please use the `y_pre` argument instead.
#> Solomon GLM (unified model)
#> Formula: y ~ treat * pretested + pre_obs
#> Covariance: HC3 heteroskedasticity-consistent; t tests (df = 90)
#>
#> Term Est (SE) t df p 95% CI
#> (Intercept) 3.019 (0.077) 38.99 90 <.001 [2.865, 3.172]
#> treat 0.110 (0.114) 0.97 90 0.336 [-0.116, 0.337]
#> pretested -1.251 (0.535) -2.34 90 0.022 [-2.313, -0.189]
#> pre_obs 0.431 (0.163) 2.65 90 0.010 [0.107, 0.754]
#> treat:pretested -0.290 (0.157) -1.84 90 0.068 [-0.602, 0.022]
#>
#> Key contrasts Est (SE) t df p 95% CI Wald R2
#> ATE (avg over pretest) -0.035 (0.079) -0.44 90 0.660 [-0.191, 0.121] 0.002
#> Pretest x Treatment -0.290 (0.157) -1.84 90 0.068 [-0.602, 0.022] 0.036
#> Treatment | pretested -0.180 (0.108) -1.66 90 0.100 [-0.395, 0.035] 0.030
#> Treatment | unpretested 0.110 (0.114) 0.97 90 0.336 [-0.116, 0.337] 0.010
#>
#> Wald R2: partial R-squared for conventional Gaussian OLS;
#> a Wald-based descriptive approximation when robust covariance is used.
# One model for all six groups.
fit_solomon_glm(post_behavior, condition, pretested, pre_behavior,
control = "Control", data = mai2020)
#> Solomon GLM (unified model), N-group design
#> Conditions: RP, GS; control: Control. With and without a pretest: 6 groups.
#> Formula: y ~ (treat_RP + treat_GS) * pretested + pre_obs
#> Covariance: HC3 heteroskedasticity-consistent; t tests (df = 126)
#>
#> Omnibus tests
#> Test Statistic p
#> Condition (avg over pretest) F(2, 126) = 1.10 0.336
#> Pretest x Condition F(2, 126) = 1.74 0.179
#> Condition | pretested F(2, 126) = 2.23 0.112
#> Condition | unpretested F(2, 126) = 0.73 0.485
#>
#> Contrasts
#> Comparison Contrast Est (SE) t df p p adj. 95% CI
#> RP vs Control ATE (avg over pretest) -0.035 (0.078) -0.44 126 0.659 0.659 [-0.189, 0.120]
#> GS vs Control ATE (avg over pretest) 0.088 (0.080) 1.10 126 0.275 0.551 [-0.071, 0.247]
#> RP vs Control Pretest x Treatment -0.290 (0.156) -1.85 126 0.066 0.133 [-0.599, 0.020]
#> GS vs Control Pretest x Treatment -0.091 (0.161) -0.57 126 0.571 0.571 [-0.409, 0.227]
#> RP vs Control Treatment | pretested -0.179 (0.107) -1.67 126 0.097 0.193 [-0.392, 0.033]
#> GS vs Control Treatment | pretested 0.042 (0.099) 0.43 126 0.670 0.670 [-0.154, 0.239]
#> RP vs Control Treatment | unpretested 0.110 (0.114) 0.97 126 0.335 0.585 [-0.115, 0.336]
#> GS vs Control Treatment | unpretested 0.134 (0.126) 1.06 126 0.293 0.585 [-0.117, 0.384]
#>
#> p adj.: adjusted by Holm's (1979) procedure within each contrast, across the 2 comparisons.
#> Confidence intervals are not adjusted.
#>
#> Experimental: in the package's simulation study (issue #45), the omnibus
#> tests of Condition | pretested and Condition | unpretested rejected in up
#> to 6.9% of replications at the .05 level with three treatments and 10
#> participants per group. No other test, and no family of adjusted
#> comparisons, failed the study's rule. See ?fit_solomon_glm.