S4 Stata verification summary
Software: Stata 16 MP
Input: S1_Dataset_SME_LSC_anonymized.xlsx, sheet "analysis_data"
Output: S4_Reproduced_Tables.xlsx and S4_Analysis_Code.log

Imported: 718 observations; 45 analysis variables
Variable-range assertions: passed
Runtime errors: none
ITCAP distribution: 417 coded 1; 301 coded 0; mean = 0.5807799
Table reconciliation: Tables 1-9 match the manuscript cell by cell
VIF diagnostic: maximum = 1.07; mean = 1.03
Table 5 check: diagonal = 1.00 without significance markers; maximum
absolute correlation = 0.21
Significance formatting: stars are attached to coefficients or AMEs only;
all reported p-value fields contain numeric p-values without stars

Table 3: KMO = 0.912; Bartlett chi2(253) = 4240.14; four factors with
eigenvalue greater than 1; first principal component = 26.94%; cumulative
variance after rotation = 46.421%. The 23-item item set is listed explicitly
in the do-file.

Pooled logit: LR chi2(6) = 54.42; Cox & Snell R2 = 0.073;
Nagelkerke R2 = 0.098; correctly classified = 445/718 = 61.98% (reported
as 62.0%). The earlier 58.77% value was the intercept-only classification
rate because the prior do-file re-estimated the null model before
estat classification.

Stage Nagelkerke R2: under 3 years = 0.075; 3 to 10 years = 0.110;
over 10 years = 0.088

Probit: McFadden R2 = 0.056; LR chi2(6) = 54.48; LogL = -459.3279;
AIC = 932.66

Lower-registered-capital model: n = 465; Nagelkerke R2 = 0.147

The do-file reproduces Tables 1-9 in the same row order as the manuscript
and writes each table to a separate sheet in S4_Reproduced_Tables.xlsx.