Does education raise productivity and wages equally? The moderating role of age and gender

Table 7 Estimates for the entire sample, three educational categories, translog specification

	GMM-SYS
	Value added per hour worked (ln)	Wage cost per hour worked (ln)	Value added-wage cost gap^c
	(1)	(2)	(3)
Lagged dependent variable (ln)	0.852^*** (0.020)	0.781^*** (0.040)	0.671^*** (0.036)
Shares of worker^a
Low-educated (E12)	Reference	Reference	Reference
Middle-educated (E34)	0.067 (0.180)	0.070 (0.015)	−0.021 (0.064)
High-educated (E567)	0.346 (0.232)	0.158 (0.138)	0.150^* (0.087)
ln capital^d	0.024 (0.040)	0.030 (0.027)	−0.095^*** (0.002)
ln capital × ln capital	0.000 (0.002)	−0.001 (0.001)	0.006^*** (0.002)
ln capital × E34	−0.000 (0.000)	0.000^* (0.000)	−0.000 (0.000)
ln capital × E567	0.000 (0.000)	−0.000^* (0.000)	0.000 (0.000)
E34 × E34	−0.080 (0.176)	−0.104 (0.103)	0.057 (0.065)
E567 × E567	−0.215 (0.224)	0.256^* (0.131)	−0.113 (0.092)
Hansen over-identification test, p value	0.607	0.680	0.258
Arellano-Bond test for AR(2), p value	0.117	0.125	0.228
Number of observations	6714	6714	6714
Number of firms	1844	1844	1844
Chi-squared statistic for equality of regression coefficients, H₀
E34 = E567	0.90	2.39	3.58^*
Interpretation^b	E12 = E34 E34 = E567 E12 = E567	E12 = E34 E12 = E567 E34 = E567	E12 < E567 E34 < E567 but E12 = E34
Interpretation^b	⇒Education has no significant effect productivity (but E567 > E12 at 14% level)	⇒Education has no significant effect on wage costs (but E567 > E34 at 12% level)	⇒E567 more profitable than E12 and E34

Notes: Standard errors, that are robust to heteroskedasticity and autocorrelation, are reported between parentheses. Regressions also control for: % of workers with 10 years of tenure or more; % workers younger than 30 and older than 49 years, respectively; % women; % part-time workers; % blue-collar workers; % workers with fixed term employment contract; % apprentices; % temporary agency workers; ln of firm size; level of collective wage bargaining; region where the firm is located (2 dummies); industries (8 dummies), and years dummies (11). AR (2) refers to second-order autocorrelation in first-differenced errors. GMM-SYS specifications include first and second lags of explanatory variables (except time dummies) as instruments
^***p < 0.01, ^**p < 0.05, ^*p < 0.1
^a Low-educated workers (E12) have at most a degree of lower secondary school. Middle-educated workers (E34) have at most a degree from upper (general, technical or professional) secondary school. High-educated workers (E567) have a tertiary educational attainment (i.e. at least a Bachelor’s or equivalent degree)
^b ‘<’ (‘=’) indicates if regression coefficients are (not) statistically different at the 10% level
^cValue added-wage cost gap = ln(value added per hour) − ln(wage cost per hour)
^dln of capital stock per worker