Efficient recoding for long measures

[Storm Hiskens]

Howdy!

I've been tasked with cleaning and recoding a dataset.

I've come across two sets of measures with 84 items. All items are stored as a strings and I am converting them to numeric for analysis.

My current code looks like this, and I will be adapting this structure code for ccnes1 to ccnes84.:

*CCNES1*
replace ccnes1 = "." if ccnes1 == ""
replace ccnes1 = "1" if ccnes1 == "(Very unlikely)1"
replace ccnes1 = "2" if ccnes1 == "2"
replace ccnes1 = "3" if ccnes1 == "3"
replace ccnes1 = "4" if ccnes1 == "(Neither likely nor unlikely)4"
replace ccnes1 = "5" if ccnes1 == "5"
replace ccnes1 = "6" if ccnes1 == "6"
replace ccnes1 = "7" if ccnes1 == "(Very likely)7"
*Converting from string to numeric
destring ccnes1, gen(ccnes1_n)
*Dropping old variable
drop ccnes1
*Renaming new variablepbace8
rename ccnes1_n ccnes1
*Labelling new variable
lab def ccnes 1 "(Very unlikely) 1" 2 "2" 3 "3" 4 "(Neither likely nor unlikely) 4" 5 "5" 6 "6" 7 "(Very likely) 7"
lab val ccnes1 ccnes
*Checking variable
codebook ccnes1

Is there a more efficient way to recode, and relabel, rather than using the above code 84 different times for the individual questions in the measure?

Thanks in advance hivemind!

[Clyde Schechter]

Yes. In fact, I see no reason to use the approach you have taken as this can be greatly streamlined with the -encode- command.

I will make the assumption that your 84 variables are all of the form ccnes*. If not, hopefully the whole list of them can be written in a few wildcard expressions that you can use instead of ccnes* in the code shown below.

The whole thing, all 84 variables, can be done in two lines of code.

lab def ccnes 1 "(Very unlikely) 1" 2 "2" 3 "3" 4 "(Neither likely nor unlikely) 4" 5 "5" 6 "6" 7 "(Very likely) 7"
encoderall ccnes*, label(ccnes)

-encoderall- is found in the -encoder- package by Daniel Klein, available from SSC.

Note: I notice that in your -replace- commands you have things like "(Very unlikely)1" with no blank space between the ) and the following number. If that is accurate, then the -lab def- command must do likewise for this code to work, the text in the value labels must exactly match the text in the original ccnes* text variables. So modify the code accordingly if that is the case. If you want the ultimate results to have a space there, then you can fix the label after the fact with

lab def ccnes 1 "(Very unlikely) 1"  4 "(Neither likely nor unlikely) 4"   7 "(Very likely) 7", modify

and the change will automatically apply to all the variables you applied the label to (or subsequently apply the label to)

Note: No example data was given, so the code is untested. Beware of typos or other errors.

[Storm Hiskens]

Much obliged Clyde! Thank you kindly