The table summarises the results from an OLS with robust standard errors, and from the following robustness checks: and OLS with primary outcome variable with rounded frequencies, a Logit regression with the primary outcome variable transformed into binary, and a double-bounded Tobit-regression
OLS with robust standard errors
OLS with robust standard errors (with rounded frequencies)
Logit with binary outcome variable
(robustness check)
Double-bounded Tobit
(robustness check)
Treatment 1
8.209*** (1.493)
8.201*** (1.487)
0.466*** (0.107)
10.51*** (1.876)
Treatment 2
11.98*** (1.494)
12.00*** (1.489)
0.677*** (0.108)
14.96*** (1.882)
Feeling of Safety 1
0.293*** (0.0247)
0.293*** (0.0246)
0.0162***(0.00160)
0.393*** (0.0283)
Agency
-0.185 (0.119)
-0.192 (0.119)
-0.0182* (0.00908)
-0.195 (0.157)
Responsible for hiring
3.897* (1.941)
4.019* (1.943)
0.294 (0.164)
4.579 (2.751)
Understanding of a risk
14.06*** (1.247)
14.06*** (1.244)
0.837*** (0.0989)
18.20*** (1.703)
Appropriateness of a risk management
0.130*** (0.0324)
0.129*** (0.0323)
0.00846*** (0.00218)
0.158*** (0.0387)
Perceived fairness of the hiring process
0.0280 (0.0301)
0.0258 (0.0298)
0.00179 (0.00212)
0.0358 (0.0375)
Knowledge of reporting channels No
-4.307* (2.081)
-4.450* (2.068)
-0.166 (0.150)
-4.745 (2.608)
Yes
2.452 (1.424)
2.512 (1.417)
0.178 (0.105)
2.840 (1.859)
Age
-0.252*** (0.0741)
-0.261*** (0.0736)
-0.0172** (0.00566)
-0.256** (0.0978)
Gender
Female
1.170 (1.778)
1.822 (1.766)
0.0964 (0.125)
2.912 (2.177)
Male
3.214 (1.844)
3.882* (1.833)
0.230 (0.129)
4.893* (2.259)
Years in the public administration
0.0867 (0.0745)
0.0860 (0.0740)
0.00620 (0.00551)
0.108 (0.0956)
Intercept
31.50***(3.917)
31.87*** (3.899)
-1.090*** (0.290)
22.78*** (5.047)
N
2537
2537
2537
2537
R2
0.248
0.250
adj. R2
0.244
0.246
var(e.likelihood_to_communicate)
1407.0*** (48.35)
Note: Standard errors in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001. Given the high concentration of observations at 50, the transformation of the primary outcome variable is expected to create noise, as values clustered around 50 will be assigned to either 0 or 100 and may this negatively affect the reliability of the results.