Table 4

Returns to Signaling: Conditional Difference‐in‐Differences

	Dependent Variable
	Project Value (Log)	Number of Projects	Number of Projects > 0	Earnings
	(1)	(2)	(3)	(4)
Treated	−0.499^***	0.021^***	0.015^***	5.081^*
	(0.083)	(0.005)	(0.003)	(2.582)
Number of microcredentials × Treated	0.081^***	0.009^***	0.006^***	3.820^***
	(0.021)	(0.002)	(0.001)	(0.784)
Number of microcredentials	−0.042^**	−0.001	−0.001	−0.220
	(0.014)	(0.001)	(0.001)	(0.148)
Fixed effects	No	No	No	No
Observations	10,805	1,501,761	1,501,761	1,501,761
Adjusted R²	0.106	0.035	0.029	0.003
Difference in pre‐trends	−0.018	0.000002	0.000002	−0.035
	(0.015)	(0.000002)	(0.000002)	(0.023)

Notes: In Column 1, the unit of observation is one project. In Columns 2–4, the unit of observation is a 14‐day pre‐ or post‐test period. In addition to the variables reported, all models include year dummies, the average rating for completed projects, a dummy variable with value 1 if a worker does not have any ratings from past projects, cumulative (arsinh‐transformed) dollars earned on the platform, and the cumulative number of completed projects on the platform, measured at the time of project start. Estimation is based on CEM matching without replacement. Variables used for matching in Column 1 are: the number of completed tests, the number of completed projects, year dummies, a no‐past‐rating dummy, and cumulative earnings and past ratings. In Columns 2–4, matching is done using the same covariates as in Column 1 and the values of the dependent variable in periods t = –1, . . ., –4. “Treated” is an indicator variable that equals 1 for each worker who belongs to the treatment group and 0 otherwise. “Difference in pre‐trends” refers to an interaction term between a linear time trend and a treated group dummy variable for the pre‐test time trend (see the text for details). Standard errors are clustered on the match group level. Significance levels in all specifications are: ^*** = 0.1%, ^** = 1%, ^* = 5%, and + = 10%