Skip to main content

Table 11 Model estimates, controlling for occupation-level physical interaction constructed using principal component analysis.

From: Lockdown stringency and employment formality: evidence from the COVID-19 pandemic in South Africa

 

Overall

Lockdown level

5

4

3

Panel A: Employment

(1)

(2)

(3)

(4)

Treatment × post

 − 0.030***

 − 0.035***

 − 0.037***

 − 0.057**

 

(0.008)

(0.013)

(0.013)

(0.027)

\({PI}_{o}^{PCA}\)

 − 0.007

 − 0.008

 − 0.018*

0.011

 

(0.007)

(0.011)

(0.010)

(0.015)

Constant

1.868***

1.440

2.039**

2.644*

 

(0.621)

(0.917)

(0.993)

(1.426)

Observations

24,678

9368

9621

5689

Panel B: Formal employment

    

Treatment × post

 − 0.006

0.002

 − 0.011

 − 0.057**

 

(0.008)

(0.012)

(0.012)

(0.026)

\({PI}_{o}^{PCA}\)

0.020***

0.026**

0.007

0.035***

 

(0.006)

(0.011)

(0.009)

(0.013)

Constant

1.547***

2.039**

0.784

1.413

 

(0.592)

(0.904)

(0.944)

(1.297)

Observations

24,678

9368

9621

5689

Panel A: Informal employment

    

Treatment × post

 − 0.028***

 − 0.035***

 − 0.034***

 − 0.004

 

(0.007)

(0.011)

(0.010)

(0.018)

\({PI}_{o}^{PCA}\)

 − 0.022***

 − 0.029***

 − 0.021***

 − 0.021

 

(0.006)

(0.009)

(0.008)

(0.013)

Constant

0.217

 − 0.833

1.325*

0.873

 

(0.487)

(0.728)

(0.776)

(1.128)

Observations

24,678

9368

9621

5689

  1. This table presents estimates of Eq. (1), overall and by lockdown level, for varying binary dependent variables while additionally controlling for occupation-level workplace physical interaction, similar to Table 7 but instead here the index is constructed using Principal Component Analysis. Sample restricted to the working-age population (15–64 years) as of 2020Q1. Lockdown levels range from 5 (most stringent) to 3 (most lenient). All models control for a vector of time-varying observable covariates including age, highest education level, and employment type, as well as individual fixed effects (FEs). PI index = Physical Interaction index generated by merging occupation-level Occupational Information Network (O*NET) data and Time Use Survey 2010 with the QLFS data. Standard errors presented in parentheses and are clustered at the panel level. Estimates weighted using sampling weights. ‘Post’ coefficient omitted for brevity
  2. ***p < 0.01, **p < 0.05, *p < 0.10