I downloaded the ACS data from 2005-2010 and hope to do some analysis on the population composition in one state, such as California. I understand that I should use “PERWT”, which is a representative of US population based on some characteristics. I was wondering what kind of characteristics this variable is based on? If I restrict my analysis to only California and count the number of people that are hispanic in California, then should the weight indicate how many hispanic in California or in the whole US? Thanks.
In general, the methodology used to construct the sampling weights in US Census and ACS data is quite complex. They help adjust for a complicated list of stratification methods designed to allow for credible analysis to be completed on sub-groups of the entire US population. More detail about the sample designs of specific IPUMS USA samples can be found here. That said, to answer your specific question, analysis at the state level using the PERWT variable will provide fairly accurate estimates. Said differently, the person level sampling weight will provide an estimate of how many individuals meet your given criteria in California. For additional detail and resources about sampling weights in IPUMS data see this blog post.