Table 3. DID estimates^a of the effect of an IEC sanitation campaign on individual household latrine ownership in Bhadrak, Orissa, India, 2005–2006.

	Model 1b	Model 2c
A: Full sample
No. of households/clusters	1050/20	2100/20
IEC effect (%)	19.0	28.7
95% CId	4.7–33.3	14.6–42.9
P-value	0.006	0.000
B: BPL only
No. of households/clusters	632/20	1264/20
IEC effect	23.7	34.2
95% CId	6.7–40.7	18.0–50.4
P-value	0.003	0.000
C: Non-BPL only
No. of households/clusters	418/20	836/20
IEC effect	12.0	20.7
95% CId	1.9–25.8	6.2–35.2
P-value	0.084	0.000

BPL, below the poverty line; CI, confidence interval; DID, difference-in-difference; IEC, information, education and communication.
^a Standard errors were corrected for clustering at the village level.
^b Estimate based on a simple comparison of means in 2006.
^c DID using observed latrine ownership in 2005 and 2006. As a check for robustness, we estimated a semi-parametric DID model²² that essentially uses inverse probability weights as a function of baseline latrine coverage, and found virtually identical results. This confirms that the combination of randomized assignment, covariate balance and DID estimation eliminates any potential bias.
^d The intracluster correlation coefficient is set at 0.125.