The only modeling I do is statistical.
Our sample comes from a frame of business establishments. This frame contains limited information, and is collected as part of the unemployment insurance system. When we sample a business, we ask for more details. Yet it is a voluntary survey.
To test out our sample design, we run a simulation. We take thousands of samples from the entire frame, and use the limited data to make simple estimates, for each sample, of the less detailed information. Since we know the same data for the entire population, we can compare each sample estimate to the true value.
Yet for our detailed survey we only
have one sample, so this simulation using less detailed information is only a model of reality.
Since our survey is voluntary, there is non-response, so to fill in the missing data, we use a model. That is, we fit a model using collected, respondent data, and then use it to fill in the missing data. We assume the missing data for a unit will be similar to units we collected that share similar characteristics, like size, industry, occupatiobs, etc.