Missing data is a pragmatic fact

that must be investigated and not

a disaster to be mitigated. It is a

property of the population to

which we seek to generalize and

can cause problems not only

through its impact on the sample

size available for analysis but

also through its potential hidden

biases. Making imputations without

first analysing the randomness

of the missing responses can

even be worse than doing nothing,

so care is needed while

imputing missing values. This

paper reflects on how to prevent,

analyse and handle missing data

and how effects of imputation can

be checked.