# Thread: Data with missing values analysis help

1. ## Data with missing values analysis help

I have data from a questionnaire asking about how they felt on a few variables (for example quality of sleep) before and after taking a fitness training course. There are 140+ responses.

However, there are several (maybe about 10) respondents who didn't answer the basic questions about their age and/or gender, although they did answer the ordinal scale questions that came later in the questionnaire (for example quality of sleep).

Would it be statistically acceptable if I keep such observations when I am analyzing only one variable at a time (for example a hypothesis test about only the variable quality of sleep)?

Could these missing values (age, gender) in the results be remedied with some imputation method, or I must delete all the observations with missing gender and/or age when I go on to jointly analyze at least one of these variables with one/several other answers in the questionnaire?

2. ## Re: Data with missing values analysis help

Hey osku809.

I took a look at my old notes for missing values [university ones] and the three classifications of missing values are missing completely at random, missing at random, and not missing at random.

Do you know about any of these?

3. ## Re: Data with missing values analysis help

chiro, can you explain me about missing random and not missing at random?

4. ## Re: Data with missing values analysis help

I dug up my notes and the definitions are as follows:

P(Ri | Y(i,O), Y(i,M), Xi) = P(Ri, Y(i,O), Xi) or that Ri is conditionally independent of Y(i,M)

Y(i,M) - Missing values [dependent variable]
Y(i,O) - Observed values [dependent variable]
Ri - Response indicator
Xi - Independent variable

The book references the following text - Applied Longitudinal Analysis by Fitzmaurice Et Al in lecture notes.