* Compare coefficients and standard errors from
*    data which is measured coarsely

set seed 1062016

postfile msd rep sample coef se using results, replace

quietly forvalues rep = 1/1000 {
	clear
	set obs 20
	generate x = runiform(-5, 5)
	generate y = 3*x + rnormal()
	regress y x
	post msd (`rep') (0) (_b[x]) (_se[x])
	generate x1 = round(x, 0.5)
	regress y x1
	post msd (`rep') (1) (_b[x]) (_se[x])
	generate x2 = round(x)
	regress y x2
	post msd (`rep') (2) (_b[x]) (_se[x])
	generate x3 = round(x,2)
	regress y x3
	post msd (`rep') (3) (_b[x]) (_se[x])
	}
	
postclose msd

use results.dta, clear
bysort sample: summarize coef se

* As we throw away information, our standard errors grow.