Date: May 3rd 2017, 1 pm - 2 pm
Place: B705


When designing a sample survey, intended response rate and sample size are two important concepts that must be taken into account. Today it is practically impossible to conduct a sample survey without nonresponse. By taking a smaller sample you may release funds to reduce nonresponse and vice versa. In order to find the optimum balance you must be able to formulate the potential response bias and the sampling error in comparable terms. We construct a random process model to be able to measure response error in terms of variance. This makes it easy to weight between potential response bias and sample size for a given cost and cost function. One result is that intended nonresponse should decrease with sample size. The model is shown to be quite robust for most nonresponse rates. We apply the technique to find situations when probability based web-panels are preferable to traditional sample surveys.​