Digital use categories and self-reported academic difficulty among graduate students in Kyrgyzstan
Description
Survey data from 95 graduate students at a public university in Kyrgyzstan, collected in April 2026. Participants rated the intensity of their use of six digital categories (social media, YouTube, messaging applications, film and series platforms, web browsing, online gaming) and their overall academic difficulty, each on a 0 to 10 single-item scale, and indicated whether they experienced unease when unable to access the internet. The deposit contains the anonymised analytic dataset, a codebook documenting every variable and all data-cleaning decisions, and the scripts used to produce all reported analyses. Direct identifiers (email addresses and telephone numbers, collected for recruitment purposes only) were removed before deposit and are not shared. Two duplicate submissions were resolved by retaining the first response, giving N = 95 from 97 raw records.
Files
Steps to reproduce
1. Download all files into a single working directory. 2. Read CODEBOOK.md first. It documents every variable, the exact wording of every item in English translation, all data-cleaning decisions, and eight interpretation notes that constrain what the data can support. Section 5.6 in particular: every construct is measured with a single item, so reliability cannot be estimated and no construct validity evidence exists for any variable in this file. 3. Load the analytic file. Variable names are in Turkish because the scripts reference them directly; English labels are in the codebook. R: d <- read.csv("analytic_dataset.csv", fileEncoding = "UTF-8-BOM") Python: d = pd.read_csv("analytic_dataset.csv") Check: nrow(d) = 95. Column means should be zorluk 5.96, sm 6.27, yt 6.03, ilet 7.83, film 3.78, web 5.34, oyun 2.06. If these do not match, the wrong file has been loaded. 4. Run 02_analysis.py (Python 3.12) or 03_analysis.R (R 4.5). Both implementations produce the same results and either reproduces every value reported in the manuscript. Console output from the Python run is provided in analysis_output.md for direct comparison. R packages: psych, car, lmtest, sandwich, lavaan Python packages: pandas, numpy, scipy, statsmodels, scikit-learn, factor_analyzer, matplotlib 5. 01_prepare_data.py documents how the analytic file was produced from the raw questionnaire export. The raw export is not included in this deposit: it contains an email address for every respondent (a required field used for recruitment) and telephone numbers, which were removed before deposit. The script is provided so that every transformation is inspectable, not so that it can be re-run. 6. Note on the three headline quantities. Academic difficulty correlates with the consumptive index at rho = .312 and with the communicative index at rho = .083; the bootstrapped difference is .229, 95% CI [.065, .397]. Parallel analysis retains a single component. If a reanalysis does not recover these, check first whether duplicate submissions were handled as described in CODEBOOK.md section 4.1.