Mousa Ajouz -

I think you would have problems with this because it sounds Iike the most important thing you are trying to measure is mostly just present in the one area in which you are actually interested and none of your data are there.

Some people have worked in nonprobability sampling using covariate data, but I would expect results here to be subject to high bias. In the following paper, reference is made to a variety of situations. See Valliant etal on nonprobability sampling in general, and Haziza etal regarding sampling tax filings from e-filings and extending this to paper filings which are not sampled. The key is good covariate data. However, that may also be hard to obtain. Also, confirming results would be problematic. Further, I expect that the more covariates you need, the less well you will be able to model, even if high quality data were available. At any rate, there might be something in the following, or it's references, which might possibly be of some help, I hope:

https://www.researchgate.net/publication/383976649_Cut-off_Cutoff_Quasi-Cutoff_and_Other_Nonprobability_Sampling_with_Inference

That paper is with regard to my presentation at the Seventh International Conference on Establishment Statistics (ICES), Glasgow, 2024.

Best Wishes - Jim Knaub

Mousa - Yes, you can use small area estimation with neighboring countries data if they share similar contexts. Ensure the model captures regional correlations and validate to avoid bias.