Читать книгу Researching in the Age of COVID-19 Vol 1 - Группа авторов - Страница 13

Among the six studies, three –​ CERRBC, GBV and a subsample of HBB –​ are follow-​up survey rounds. Follow-​up survey rounds have a significantly higher response rate compared to initial survey rounds. Table 1.4 presents the regression results.

Table 1.4 Regression: follow-​up survey

	(1)	(2)	(3)	(4)
	OLS completed survey	FE completed survey	OLS calls	FE calls
Follow-​up = 1, follow-​up survey round	0.372***	0.256***	-​0.136*	-​0.332
	(0.0142)	(0.0762)	(0.0820)	(0.385)
Constant	0.373***	0.389***	3.152***	3.178***
	(0.00569)	(0.0115)	(0.0317)	(0.0594)
Observations	8,356	8,356	8,356	8,356
R-​squared	0.066	0.110	0.000	0.074
Study FE		Yes		Yes

Notes:

Robust standard errors in parentheses. *** p < 0.01; ** p < 0.05; * p < 0.1.

In the first two models, the dependent variable is ‘completed survey’ (1 = completed, 0 = not completed). In models (1) and (2), the authors regress a binary variable (1 = follow-​up, 0 = first survey round) and observe that follow-​up survey rounds have 25.6 per cent higher response rates (significant at p = 0.01 levels) compared to first survey rounds. Follow-​up survey rounds use 13.6 per cent fewer calls to complete a survey (significant at p = 0.1 levels), though this does not hold true when study fixed effects are introduced.

Finally, preliminary results from the study show that the incentive structure (flat versus dynamic) and SMS-​based scheduling attempts have no effect on response rates. These findings could benefit from further validation from other studies conducted in different contexts in order to become more generalizable.