Using Survival Analysis To Predict Sample Retention Rates

Andy Sadler and Larry Lang


The International Price Program (IPP) publishes monthly price indexes which measure the average change over time in prices of internationally traded products and services. IPP's sampling methodology supports its published merchandise price indexes by attempting to maintain a minimum number of items, within each published stratum, over the stratum's two year life cycle. Stratum retention rates — formerly calculated by using an averaging process — are used to set minimum item allocations. This paper uses the Life-Table estimate of the survivor function applied to censored data to model retention rates. Our conclusions are: (1) The model produces a markedly superior fit to that produced from using simple averages; (2) The model is un-affected when samples are rotated; (3) The model allows us to use the Lower Confidence Limit which gives conservative estimates of the survival rates