1
$\begingroup$

I have a customer satisfaction survey with around 150 responses and six questions, and asking for gender, age, education, and place of residence.

Around 130+ have answered every question in the survey. But for the rest, there is at least one question that each has not answered, and some have multiple questions not answered.

For example, the missing values might look like this:

  1. Did not state one or more of gender, age, education, or place of residence, but answered all else.

  2. Did not state more than one or any of these four characteristics.

  3. Did not answer one of the six questions, for example, "How likely would you recommend our offering to someone else?"

  4. Did not answer many of the six questions.

  5. At least did not answer one of the 4 characteristics and at least did not answer one of the six questions.

  6. Did not answer many or any of the 4 characteristics and did not answer many of the six questions.

So what should I do now with these respondents with missing values? Can I still use the answers by respondents who skipped at least one item (a characteristic or question) for descriptive statistics?

What should I do if I want to proceed to some hypothesis testing? Would it be a good idea to delete all the respondents with missing values altogether for this purpose? Or what do you suggest?

Improve this question
$\endgroup$
4
  • $\begingroup$ try multiple imputation? $\endgroup$
    – kjetil b halvorsen ♦
    Commented Mar 31, 2017 at 17:53
  • $\begingroup$ Although I find it well-written, this is too broad to be manageably answered here. Approaches to handling missing data are the subjects of entire books and courses. I wish you luck, though ~ $\endgroup$
    – rolando2
    Commented Mar 31, 2017 at 19:40
  • $\begingroup$ This has been covered many times on this site. $\endgroup$
    – Michael R. Chernick
    Commented Apr 1, 2017 at 0:08
  • $\begingroup$ Possible duplicate of Dealing with missing values where the question was not asked $\endgroup$
    – Michael R. Chernick
    Commented Apr 1, 2017 at 0:08

1 Answer 1

Reset to default
0
$\begingroup$

Deleting these rows of data generally means you are assuming that these missing data values have no significant relationship with the your survey.

In real world research, missing values are usually not missing completely at random. There is a likely reason which is linked to your survey itself.

Another disadvantage of deletion is that will reduce your sample size which will not help with your hypothesis testing and also you may end up producing biased parameter estimates.

One of the more robust methods to deal with missing values is Multiple Imputation(http://www.stefvanbuuren.nl/mi/mi.html). You could even try forms of regression imputation which would tell you a relationship between the variables in your dataset and the missing data.

This way you are substituting the missing values with values that help you minimize the loss of information due to missing data

Improve this answer
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.