It appears that your reference dataset includes observations for statename/year combinations that do not appear in your master dataset. That's not unusual when combining data from two different sources. You can drop the observations from the using dataset that did not have a match in the master dataset, and also drop the _merge variable that is no longer needed, with the following code.
Code:
merge m:1 statename year using "faminc_smooth2.dta" // the next command should show 26 observations removed from your dataset drop if _merge==2 // the next command will remove the variable _merge from your dataset drop _merge
Comment