How to approach missing data...ignore or not to ignore? What would be the best option when you come across missing data. Do you exclude that person completely from the data analysis or not. If not then how do you go about it?
i have data on two groups at three different time points. At each time point they had to undertake a 6 minute walking test. Now some participants they stopped at stage 4 (because physically they could not go any longer) and others managed to complete all of the stages (1-6th stage)..some managed to complete only 2 stages and so on. 
Thank you in advance
 A: The question posed is interesting, although it is hard to give an exact answer. 
I'll go through some of the common concerns below.
When do missing data pose a problem?
In general, missing data poses a problem when it is not at least missing-at-random (MAR). Such data may lead to severely biased estimates of population quantities. In your case, the persons who did not show up for the follow-up tests may differ from those who did in important ways.
However, it is important to note that missing data (esp. in the form of "Don't know"-answers, common in attitude surveys) is not intrinsically uninformative. Schumann and Presser (1996, ch. 4) has a good walk-through of what one must consider when dealing with "don't know"-answers.
What can be done about missing data?
There are two main approaches to missing data: deletion and imputation. I'll summarize them briefly below.
Case-wise deletion is the omission of observations which are missing in one or more of the variables of interest. It is my belief that this is the standard in SPSS. Pair-wise deletion, on the other hand, computes the relevant statistics on pairs of variables. If you were interested in the correlations between $X,Y,Z$ and you had an observation missing in $Z$, but valid in $Y,X$, case-wise deletion would drop if from the analysis entirely, while pair-wise deletion would only drop it from the analysis of $corr(X,Z),corr(Y,Z)$.
Both deletion methods have their weakness: case-wise deletion often leads to a low sample size (and thus, low power) while pair-wise deletion obscures the size of the sample and - in my opinion - complicates inference about the data generating mechanism.
Imputation refers to the attempt to impute missing values. Once, the standard approach was so-called mean imputation, where one substitutes the mean of the valid observations in the missing observations. Newer development has lead to more efficient and useful techniques, which Wiki has a good summary of.
A: If your dependent variable is number of stages walked, the missing data problem disappears.
The analysis could then be a multilevel ordinal logistic model. Unfortunately, the two programs you list in your tags SPSS and Stata are two that I am not familiar with, even to knowing whether this type of analysis is possible in those programs. It can be done in R with the nlme package and in SAS with PROC GLIMMIX. 
If your dependent variable is some score measured at each time point (such as heart rate) then you may have a big problem because the data may be severely missing not at random. It depends on whether the differences between the people who stop at an early stage and those who finish can be accounted for by other variables in your data set (that is, are they missing at random)? This is not a fully testable assumption. If you have a lot of measures on the subjects (age, weight, BMI, resting heart rate, exercise pattern etc) that are related to the number of stages completed, it may be reasonable. 
