Table 3:

Behavior Change After Inferred Diabetes Diagnosis

	Calories			Expenditures	Expenditures, Magnet HH
	Prob. Interaction	All TS Purchase	Predict Diab >=50%	Prob. Interaction	Prob. Interaction
	(1)	(2)	(3)	(4)	(5)
Month Before, Month After	−3019.6***	−590.3**	−1797.5**	−2.37	−6.52
	(937.5)	(296.9)	(828.2)	(2.49)	(3.97)
		[−2951.5]	[−2869.9]
2–12 Months After	−1189.3	216.7	−468.6	−0.40	−4.41
	(841.1)	(321.0)	(914.5)	(2.54)	(4.04)
		[1083.8]	[−748.2]
Household Fixed Effects	YES	YES	YES	YES	YES
Year-Month FE	YES	YES	YES	YES	YES
R-squared	0.49	0.49	0.51	0.66	0.675
Number of Obs	112,416	112,416	13,896	112,416	64,680
Number of HHs	4684	4684	579	4684	2695

Notes: This table shows the evidence on calories and expenditure changes. The omitted category is the year before diagnosis, not including the month before. In Columns (1), (5) and (6) the independent variables are timing interacted with the probability of being diabetic, as inferred from the machine learning approach. These coefficients can be interpreted as the impact of diagnosis. In Columns (2) and (3) the independent variables are timing alone, and the sample is either the overall sample, or limited to households with higher predicted probability. Standard errors are in parentheses. The figures in square brackets in Columns (2) and (3) show the estimated impact scaled up to 100% diagnosis. This should be comparable with the coefficient in Column (1). Regressions include controls for household fixed effects and year-month fixed effects. Magnet households are those who also scan and report prices for non-UPC coded goods

^*significant at 10% level

^**significant at 5% level

^***significant at 1% level.