Low-carbon policy and employment: heterogeneity of workers with different skills

Abstract

This study aims to reveal the effects of low-carbon policy on labor employment structures utilizing a dataset from 295 cities in China from 2005 to 2020. The difference-in-difference design was employed to analyze and verify the mechanism. Results indicate that implementing urban low-carbon transition policies significantly reduces employment opportunities for low-skilled workers while promoting employment opportunities for high-skilled workers. Furthermore, the scale of urban employment was found to have increased considerably. The influence of urban low-carbon construction on skill structure varies depending on regional characteristics and industrial heterogeneity. Notably, green technology innovation and industrial structure transformation can positively impact high-skilled workers while negatively affecting low-skilled workers. The employment effects of low-carbon policy are supported by a range of robust approaches. Governments should lead to employment policies that are compatible with low-carbon policies, including livelihood security for workers and optimized supply of skilled labor.

Appendix A. Goodman-Bacon decomposition of the treatment effects

As argued by Goodman-Bacon (2021), when the treatment timing is varied across treated units, the two-way fixed effects DID estimator is a weighted average of all possible 2 × 2 DID estimators that compare timing groups to each other. In other words, the two-way fixed-effects DID estimate is the sum of the products of the DID estimates and their comparison weights. Possible 2 × 2 combinations are late treatment group vs. early treatment group, early treatment group vs. late treatment group, no treatment group vs. late treatment group, and no treatment group vs. early treatment group.

Assume a balanced panel with T periods (t) and N cross-sectional observations (i). The panel is divided into three groups depending on the treatment time: the early treatment group k, who receive the treatment at the point t_i = k; the late treatment group l, who receive the treatment at the point t = l > k; and the untreated group U, whose treatment time can be denoted as t_i = $+ \infty$.

For such three groups, we can divide the discussion into four scenarios. The first is the early treatment group k with the untreated group U. The second is the late treatment group l with the untreated group U. The third is the early treatment group k with the late treatment group l. The last group can be further divided according to the time treatment difference, the late treatment group, and the early treatment group as the control group. The grou** are shown in the following equations:

$$\widehat{\beta }_{jU}^{2 \times 2} = (\overline{y}_{j}^{POST(j)} - \overline{y}_{j}^{PRE(j)} ) - (\overline{y}_{U}^{POST(j)} - \overline{y}_{U}^{PRE(j)} ),j = k,l$$

(3)

$$\widehat{\beta }_{kl}^{2 \times 2,k} = (\overline{y}_{k}^{MID(k,l)} - \overline{y}_{k}^{PRE(k)} ) - (\overline{y}_{l}^{MID(k,l)} - \overline{y}_{l}^{PRE(k)} )$$

(4)

$$\widehat{\beta }_{kl}^{2 \times 2,l} = (\overline{y}_{l}^{POST(l)} - \overline{y}_{l}^{MID(k,l)} ) - (\overline{y}_{k}^{POST(l)} - \overline{y}_{k}^{MID(k,l)} )$$

(5)

where MID(k, l) indicates the time between time points k, l for the latter two equations to make comparisons (there is always one group that is treated and another that is not).

The DD decomposition theorem is described in detail next. Assuming that the dataset has $k = 1,2, \ldots K$ groups sorted according to the time point $k \in \left( {1,T} \right]$ at which they received the treatment, and that there is perhaps a group U that never received the treatment, the OLS estimate ${\widehat{\beta }}^{DD}$ is a weighted average of the following possible 2 × 2 DID estimates, as shown in Eq. 6. Equation 6 provides a complete description of the sources of identified variation and their significance in the two-way fixed-effects difference-in-difference (TWFEDD) estimator.

$${\widehat{\beta }}^{DD}={{\sum }_{k\ne U}{s}_{kU}\widehat{\beta }}_{kU}^{2\times 2}+{\sum }_{k\ne U}{\sum }_{l>k}\left[{s}_{kl}^{k}{\widehat{\beta }}_{kl}^{2\times 2,k}+{s}_{kl}^{l}{\widehat{\beta }}_{kl}^{2\times 2,l}\right]$$

(6)

In addition, the average treatment effect received by group k over period W is denoted as Eq. (7). where W in time usually represents the post-treatment period in the 2 × 2 form.

$$ATT_{k} (W) = \frac{1}{{T_{w} }}\sum\limits_{t \in W} {E\left[ {Y_{it} (k) - Y_{it} (0)\left| {t_{i} } \right. = k} \right]}$$

(7)

The total ATT is all 2 × 2 DID estimates in the sample. All 2 × 2 DUDs can be roughly classified into three categories: (i) the group treated later served as the control group for the group treated earlier (i.e., Earlier Treatment vs. Later Control), (ii) the earlier treated group served as a control for the later treated group (i.e., Later Treatment vs. Earlier Control), and (iii) Never Treated vs. Treatment.

The Goodman-Bacon decomposition results are shown in Table 10. The table presents contains the weighted average of DID estimates, each DID estimate, and the weights of the corresponding comparisons from the three estimations. Table 10 Panels A, B, and C show the TWFEDD estimates coefficients of the impact of the Low Carbon City Pilot policy on low-skilled employment, high-skilled employment, and labor force size, respectively. Table A1 shows that the weighted average of the DID estimates is similar to the coefficients for Pilotcity in Table 2.

Table 10 Results of the Goodman-Bacon decomposition