Imputation for Industrial Survey What do you think the best imputation method is for industrial surveys? 
Methodology reports usually mention historical background, administrative data or hot-deck, but according to the theory, the method of maximum likelihood or imputation multiple seems to be the best. What do you recommend me?
 A: This is a partial answer since the question is wa-a-a-ay too broad.
Maximum likelihood is an estimation method, not an imputation method, although some variations can be applied to missing data problems. It does not naturally produce imputations of any kind.
Survey statisticians are usually very cautious working with heavily-model-dependent methods, such as any GLM-type models or even MI, as far as establishment surveys are concerned, because the data are usually very highly skewed. So on one hand, transformations towards normality seem like a knee-jerk reaction, but on the other hand, the imputed data have to satisfy a lot of accounting constraints. E.g., total revenue must equal the sum of total compensation, total cost of inputs, net interest payments, and profit... and I am sure I am forgetting something since I am not an accountant. If one or more of these is imputed from say a lognormal model, there is no guarantee that the accounting inequalities would hold.
A recent handbook on business surveys has a chapter on editing and imputation that may help.
