Canadian Community Health Survey 1.1 Question

Question

I have been using the CCHS data file (cycle 1.1) PUMF and the dictionary for the PUMF. While the dictionary often lists "don't know," "refusal," "not stated" categories the actual PUMF file only lists missing (never not stated, etc.) Is it okay to assume that all those missing are either: not applicable, don't know, refusal or not stated?

Answers

1. The CCHS 1.1 PUMF does include the range of values for instances of missing information: not applicable, don't know, refusal or not stated. We have a version of this file in SPSS that includes these values and that declares them individually as missing.

Is your patron working with SPSS or SAS? I'm guessing that she or he is using SAS. Why SAS? Because of the way users typically assign missing data in this statistical system. For example, the variable CCCA_131 records whether the respondent has cancer. The values and labels for this variable are:

1 = YES
2 = NO
7 = DON'T KNOW
8 = REFUSAL
9 = NOT STATED

I have seen many researchers working in SAS assign missing data
using the following statement in a DATA step:

if ccca_131>2 then ccca_131=.; which treats the values 7, 8 and 9 as one missing category.

SAS does allow specifying 27 special missing values by using a decimal point followed by a letter or the underscore character. Therefore, a researcher could declare each of the missing values for CCCA_131 in three statements:

if ccca_131=7 then ccca_131=.A;
if ccca_131=8 then ccca_131=.B;
if ccca_131=9 then ccca_131=.C;

All three of these values would be treated as missing but SAS would differentiate between the three types of missing information. I haven't seen many researchers take the time to write this much code for all of the variables they are using, though.

2. The most recent versions of Stata includes support for multiple missing values, but earlier versions (before Stata) did not. So if your user converted the file using a program such as DBMSCopy, which converts to an older Stata format, then all of the different SPSS missing values would have been mapped onto a single Stata missing value. This might also happen if she coverted the file to the current Stata version, but used a program to convert that didn't deal well with either SPSS or Stata missing cvalues.

One solution might be to open the file in SPSS, undeclare the missing values, then do the conversion. After this was done the user could then declare the Stata missing values from within Stata.

3. Stat-Transfer, another data conversion program claims to handle multiple missing values. On the first options tab, there's a set of options for handling user-declared missing values (such as SPSS has), one of which is to convert them into Stata extended missing values.

However, I ran in to problems trying a simple transfer from SPSS using a dataset I had on hand - it took three different missing values on one variable, and returned a variable with only two - it combined two of the different values into a single Stata missing value. I spoke with a colleague who say that he's had the same problem.

Another option for handling user-declared missing values in StatTransfer is to "Use none" - this option simply preserves the original values of the variable. On the quick test I ran, this actually worked, and it's quicker than opening the file in SPSS and redeclaring all the missing values separately.