Sample design of panel surveys from the perspective of both survey errors and survey costs: the case of longitudinal social protection surveys in Latin America
Jul 25, 15:45
The impact that decisions taken in the design of longitudinal surveys have in the Total Survey Error has not been sufficiently studied yet (Lynn & Lugtig, 2017). Deciding if the longitudinal study is fixed-timed or open-ended, who to follow in time, how long should be the period between one round and the other are design decisions that have a direct influence on the quality and costs of a longitudinal survey. The assessment of the trade-off between the costs of a panel survey and the control of errors such as sampling error, nonresponse error, measurement error and processing error is key knowledge when designing a longitudinal survey and estimating its costs for the whole time-frame defined for the survey.
This research seeks to develop some general principles about the relationship between different longitudinal survey designs, some components of the Survey Total Error and the costs associated to different scenarios generated through micro-simulations.
I’ll work with the databases of the first wave of the Longitudinal Social Protection Survey (LSPS) that has been developed in four countries in Latin America and the Caribbean. I’ll also work with the information of the costs structure of the survey’s fieldwork of each of the four countries.
The survey designs that I’ll study are: fixed panel with refreshment, rotating panel and split panel. The components of the Total Survey Error to evaluate from the perspective of costs and design are: sampling error (considering different sample sizes and design effect), nonresponse error and adjustment error.
This research is based on the experience obtained from a consultancy that I’ve done for Inter-American Development Bank (IADB) called: "Sample designs in time: a proposal for the LSPS". The context of this consultancy work is that between 2012 and 2015, Colombia, Uruguay, Paraguay and El Salvador made their first LSPS measurement. The sample was around 15,000 individuals in El Salvador, Colombia and Paraguay and 18,428 individuals in Uruguay. Currently the most pressing issue of the LSPS has to do with its continuity given the economic costs of the current design. Even though the four countries value the longitudinal approach to labor and social security provided by LSPS, it is considered unaffordable with its current design. In this scenario, the specific request from IADB for my consultancy was to propose methodological changes to the initial design of the LSPS to assure its continuity and sustainability maintaining the main objectives of the LSPS.