1[2 {"id":"C1","text":"The unadjusted association between training hours and next-year productivity growth is negative (about -0.13 pp/hour, 95% CI excludes zero).","results_keys":["naive_estimate","naive_ci_low","naive_ci_high"],"status":"supported","caveat":"Association only; confounded by baseline skill."},3 {"id":"C2","text":"Adjusting for baseline skill_index, one additional training hour per employee raises productivity growth by about 0.25 pp (HC1 95% CI 0.234 to 0.273).","results_keys":["effect_estimate","effect_se","effect_ci_low","effect_ci_high","conclusion"],"status":"supported","caveat":"Causal only under selection on observables given skill_index alone, linear homogeneous effect; untestable."},4 {"id":"C3","text":"Baseline skill is a strong negative predictor of growth, which explains the sign reversal between the naive and adjusted estimates.","results_keys":["skill_index_coefficient","corr_skill_training"],"status":"supported","caveat":"Coefficient on a control has no causal interpretation by itself."},5 {"id":"C4","text":"The adjusted model fits much better than the unadjusted one (residual RMSE 1.98 vs 3.86 pp).","results_keys":["model_metric","baseline_metric"],"status":"supported","caveat":"In-sample fit; better fit does not establish identification."},6 {"id":"C5","text":"Excluding the 12 rows with negative training hours does not materially change the results.","results_keys":["sensitivity_all_rows_effect_estimate","sensitivity_all_rows_naive_estimate","effect_estimate","n_analysed"],"status":"supported","caveat":"Undetectable errors in positive values remain possible."}7]8