There are a few ways to do this. (A) A full SQUEAC. This will give you a lot of information on barriers to coverage as well as a coverage estimate. (B) A plain CSAS. You'd probably want to aim for a sample size suitable to estimating a proportion of 70% with a useful precision (i.e. +/- 10 or better). For this you'd need a sample size of 80 or more. (C) A Bayesian CSAS. This is as above but you can factor in what you already know about coverage from your previous CSAS, routine monitoring data, &c. to give a prior. You would combine this prior with new CSAS data using a beta-binomial conjugate analysis. The advantage of this approach is that you would need a smaller sample size as the prior contributes information. If (e.g.) you had a reasonable believe that coverage was most likely to be about 70% and very unlikely to be below 50% and extremely unlikely to be above 90% then a sample size of n = 42 might suffice to give you 10% precision whilst retaining some protection against bias from a misspecified prior. Details of this approach can be found in the [url=http://www.fantaproject.org/monitoring-and-evaluation/squeac-sleac]SQUEAC / SLEAC technical reference[/url] and software is available [url=http://www.brixtonhealth.com/bayessqueac.html]here[/url]. (D) Use SLEAC. This is a cut-down CSAS type survey that provides a classification. A typical sample size is n = 40. Details of this approach can be found in the [url=http://www.fantaproject.org/monitoring-and-evaluation/squeac-sleac]SQUEAC / SLEAC technical reference[/url]. I hope this is of some use.