Why people leave — burnout between workload and turnover intent

Raw data 기제_업무량_소진_이직의도.csv · 200 rows

Synthetic practice data — not a real survey and not any organization’s data. A seeded generator in the repository builds the 200-row sample with a workload → burnout → turnover-intent path planted as coefficients, so the indirect effect is really there (the planted coefficients are written in that generator). Planted values and measured values differ: the generator also planted a small direct effect, but in this sample the engine did not find it significant — the verdict this sample shows is the engine’s. Every table and APA sentence is a measured result produced by running the repository’s own engines.

The study, step by step — what was actually produced

1. ① The shape — all five columns at once

   Mean, standard deviation, range, skewness and kurtosis for the two controls (tenure, age) and for cause, mediator and outcome. Look at how each variable is shaped before asking about mechanism — a variable piled up on one side changes how the later coefficients read.

   Descriptive statistics were computed for tenure_years and 4 other variables (N = 200). For tenure_years, M = 13.09, SD = 7.11, Mdn = 13.50, range = 1.00–25.00. Skewness = 0.01 and kurtosis = -1.20, within normal range. For age, M = 41.78, SD = 10.30, Mdn = 42.00, range = 24.00–58.00. Skewness = -0.10 and kurtosis = -1.28, within normal range. For workload, M = 3.08, SD = 0.61, Mdn = 3.10, range = 1.40–4.60. Skewness = -0.13 and kurtosis = 0.27, within normal range. For burnout, M = 2.69, SD = 0.59, Mdn = 2.70, range = 1.00–4.10. Skewness = -0.01 and kurtosis = -0.42, within normal range. For turnover_intent, M = 2.45, SD = 0.69, Mdn = 2.45, range = 1.00–4.50. Skewness = 0.04 and kurtosis = -0.49, within normal range. No missing data were present.

2. ② Correlation — what moves together

   Cause, mediator and outcome make three pairs. What you can see here stops at “they move together” — a correlation matrix cannot say what travels through what. That is why the next step exists.

   A Pearson correlation analysis was conducted (N = 200). burnout and workload were significantly correlated in a positive direction, r = .59, 95% CI [.49, .67], p < .001. turnover_intent and workload were significantly correlated in a positive direction, r = .42, 95% CI [.30, .53], p < .001. turnover_intent and burnout were significantly correlated in a positive direction, r = .65, 95% CI [.57, .73], p < .001.

3. ③ Mediation — does it travel through burnout

   Does workload (X) reach turnover intent (Y) by way of burnout (M)? Baron and Kenny’s three steps come with a bootstrap confidence interval, and the decision sits on whether the indirect effect’s interval contains zero. Full or partial mediation is the engine’s verdict too — this sample carries that verdict as it came.

   Baron & Kenny (1986) mediation analysis with bootstrap testing was conducted (N = 200) to examine whether burnout mediates the effect of workload on turnover_intent. A mediation analysis examined whether burnout mediates the relationship between workload and turnover_intent. The total effect of workload on turnover_intent was significant (c = 0.478, p < .001). The direct effect, controlling for burnout, was not significant (c′ = 0.069, p = .357). The indirect effect (a×b) was 0.408, with a bootstrap 95% CI [0.311, 0.509] based on 5000 resamples. As the CI did not include zero, the indirect effect was significant, indicating full mediation.

4. ④ Hierarchical regression — how much each block adds

   Enter the two controls first, then the cause, then the mediator. Each block reports the gain in explained variance and whether that gain is significant. It is the same question wearing another face: mediation asks whether the path exists, this asks how much explanatory power the path brings. VIF comes with it, answering whether the predictors are too much alike.

   A hierarchical regression was conducted to predict turnover_intent (N = 200). In Step 1, tenure_years, age were added. The model explained 8.5% of variance, R² = 0.0845, adjusted R² = 0.0752. In Step 2, workload were added. The model explained 25.3% of variance, R² = 0.2531, adjusted R² = 0.2416, ΔR² = 0.1685, ΔF(1, 196) = 44.228, p < .001. In Step 3, burnout were added. The model explained 45.9% of variance, R² = 0.4589, adjusted R² = 0.4478, ΔR² = 0.2059, ΔF(1, 195) = 74.196, p < .001. age was a significant predictor of turnover_intent (B = 0.012, β = 0.179, t = 3.313, p = .001). burnout was a significant predictor of turnover_intent (B = 0.669, β = 0.573, t = 8.614, p < .001). Multicollinearity was assessed using the variance inflation factor; the maximum VIF was 1.62 (workload, based on final-step predictors), indicating no serious multicollinearity.

Key results

Representative quotes

Every quote is a sentence that exists in the raw data (excerpt-existence check passed).

Paper-ready tables