Statistical validity of a survey (the data is a survey, not a poll) is based on making sure your sample size is large enough and demographically representative to look at a representative smaller population than the whole of a community to have good quality data. For a simple example - you examine a subset of all the parts in a lot or of the lots in a production run, not the whole production run. No need to inspect every piece as long as you sample enough to establish a statistically valid result. Things are more complex outside of simple QC. Generally the margin of error of a survey improves as the survey group number goes up, 1/√N, but there's no need to involve the total population you're trying to represent if your group sampled is 10,000 or more. That gives you a margin of error of 1% in reliability of the data. 33,000 out of a population of 37,000,000 gives a 0.5% margin of error for a 95% confidence level in the data. Make it harder and drive the confidence level to 99% for the same census and sample group and you get a 0.7% margin of error. Pretty much anything above 10,000 is a good sample group size. 100 level statistics (I hated taking the class at the time, but it has been helpful).