Missing values

[Klaudia Lennerling]

Hello, fellow stata lovers!

I am working on my thesis and have a dataset from the Enterprise Survey available with 241 observations. However, the observations consist of 50 missing values for my dependent variable and two missing values in one of my control variables. I have assumed that they are missing completely at random since the observations with missing values do not systematically differ from the other observations. Hence I have chosen to ignore the missing values.
Is this justified? Or should I handle them in a different matter?

Best,
Klaudia

[Fei Wang]

If you have checked that all the other variables do not significantly differ between the missing and non-missing groups, then I think you could argue analyses based on the remaining observations still represent the same underlying population. Moreover, you could try multiple imputation (-mi-), as ignoring the missing may lead to a substantial efficiency loss given the limited size of sample.

[Carlo Lazzaro]

Klaudia:
if the probability of being missing is unrelated to the observed data, you have a missing completely at random (MCAR) mechanism. If that were the case with your data, set aside a loss of efficiency due to drop in sample size (that may well be remarkable, as Fei wisely pointed out), the observed data can be considered as a random subsample of the original sample.
Conversely, if the probability of missingness is the same only within groups defined by the observed data, you're dealing with missing at random (MAR) data (see https://stefvanbuuren.name/fimd/, page 8).
MAR mechanism, which is more frequent and realistic than MCAR requires -mi-.

[Klaudia Lennerling]

Hello!

Thank you for your quick response kind Fei and Carlo. MI is sadly not possible for our dataset. However, when I say ignore I do not mean dropping them from the dataset instead I have chosen to still include these observations in my regressions since as you say the probability of being missing is unrelated to observed data.

Is still including these observation a problem?

Best regards,
Klaudia

[Carlo Lazzaro]

Klaudia:
no, it isn't, as Stata will omit by default each observation with missing values in any variable (listwise deletion).
A correcty strategy for dealing with missing values is relevant especially if you're planning to submit an excerpt of your thesis to aa target journal of your research field (nowadays, reviewers and editors are more sensitive to missing values than they were in the past).
That said, since you're drafting your thesis, my advice is to negotiate each and every step of your quantitative analysis with your supervisor/teacher/professor and receive her/his clerance on that (preferably in written form), in order to avoid hard landings when the runway is in sight!

[daniel klein]

MI is sadly not possible for our dataset.

Why not?

[Klaudia Lennerling]

Klaudia:
no, it isn't, as Stata will omit by default each observation with missing values in any variable (listwise deletion).
A correcty strategy for dealing with missing values is relevant especially if you're planning to submit an excerpt of your thesis to aa target journal of your research field (nowadays, reviewers and editors are more sensitive to missing values than they were in the past).
That said, since you're drafting your thesis, my advice is to negotiate each and every step of your quantitative analysis with your supervisor/teacher/professor and receive her/his clerance on that (preferably in written form), in order to avoid hard landings when the runway is in sight!

Thank you very much Carlo for your answers, we will take your advice and discuss is more thoroughly with our supervisor.

best regards,
Klaudia