The Data Missing Mechanism And Its Testing

Parameter estimation or statistical analysis with missing data in a realisticproblem, is usually a complicated work in this situation, because it involves themissing mechanism of the data. In the normal regulations processing, the assumptionof the missing mechanism of data often is missing at random(MAR) or missing atcompletely random(MCAR), but take no account of not missing at random. Butwhether this kind of assumption establish or not, no one know it. This paper used akind of a simulation method to test missing mechanism.This paper firstly introduces the concept of the missing mechanism andillustrates the probability form, and introduces the development background of thelogit model; secondly, gives a parameter estimation of linear regression model underMAR and NMAR mechanisms. The result shows that the different missingmechanism effects the result of parameter estimation. So under the NMARmechanism, parameter estimation used the MAR mechanism is bias. Consequently,testing the missing mechanism before doing a parameter estimate is necessary verymuch; Finally, with the validity of the logit model for missing data, test the missingmechanism can be converted to get a parameter estimation of a logit model. Afterchecking the significance of the coefficient in logit model, we can give the hypothesistest of different missing mechanism. Three examples are given to demonstrate themethods under different hypothesis testing, meanwhile the examples selvesdemonstrate this arithmetic is valid.