Measurement errors in consumption surveys and the estimation of poverty and inequality indices

This paper firstly aims to evaluate the incidence of measurement error affecting the main variables collected in surveys on consumption. The assessment is carried out on two Tanzania surveys which provide both diary and panel data. Diary data can be employed to obtain reliability coefficients for time-invariant variables. When variables vary over time, as in the case of panel data, an estimation of the incidence of measurement error on the total variance can be obtained by applying models which allow the decomposition of observed variability into true dynamics and noise (e.g. the Heise model and the latent Markov model). Some evaluations of the reliability of the data are also conducted on the basis of the internal consistency criterion, an approach that does not require panel data. On the basis of the reliability estimates obtained, examples of possible impacts of measurement errors on poverty analysis are briefly discussed. These experiments clearly show the importance of the topic in poverty and inequality data analysis.