@mastersthesis{bal2025literary,
  author = {Bal, Onur},
  title = {Literary Work Readership and Its Relationship with Mentalisation and Empathy},
  school = {Yeditepe University},
  year = {2025},
  type = {Master's thesis},
  url = {https://onurbal.com/en/academics/thesis/}
}

3.4. Data Analysis

All analyses were conducted using Python (version 3.13.3; Python Software Foundation, 2025) on macOS Sonoma (version 14.4.1). The following open-source packages were used: pandas (version 2.1.2) for data structures and data manipulation (McKinney, 2010), NumPy (version 1.26.1) for numerical operations (Harris et al., 2020), Pingouin (version 0.5.5) for correlation analysis (Vallat, 2018), scikit-learn (version 1.3.2) for standardisation of continuous variables (Pedregosa et al., 2011), and Statsmodels (version 0.14.0) for linear and hierarchical regression modelling (Seabold & Perktold, 2010). Visualisations were created using Matplotlib (version 3.8.1; Hunter, 2007) and Seaborn (0.13.0; Waskom, 2021).

All participants (N = 258) completed every measure in full. No missing data imputation or listwise deletion was necessary. Logarithmic values for participants’ age and number of books read were used in correlation and regression analyses. Correlational analyses were conducted using Pearson’s r, and hierarchical regression models were built to examine the effects of demographic and psychological variables on empathy and mentalisation. Hierarchical multiple regression analyses were conducted using ordinary least squares (OLS) estimation with classical (non-robust) standard errors. Variables were entered in three blocks: demographic control variables, primary predictors, and interaction terms. Categorical variables in the control block were dummy-coded using reference categories. Predictors involved in interaction terms were mean-centred prior to computing product terms. Standardised beta coefficients (β) were calculated by multiplying each unstandardised coefficient by the ratio of the standard deviation of the predictor to that of the outcome (i.e., β = B × SDₓ / SDᵧ). Model assumptions, including normality of residuals, homoscedasticity, multicollinearity (VIF), and influence (Cook’s distance), were checked and met. Results were reported in accordance with APA 7 guidelines (American Psychological Association, 2020), with unstandardised coefficients (B), standard errors (SE), standardised beta (β), t-values, p-values, and model statistics (ΔR², R², F_change, p_change and total model F) for each step. An alpha level of .05 was used for significance testing in regression analyses, while both .05 and .001 thresholds were considered when interpreting correlation results.

All materials, including the dataset, analysis scripts, and survey preview, are available in an OSF repository (see Appendix L).