OBJECTIVES:  The socioeconomic status (SES) of households is changing rapidly all over the world, especially for middle-income and developing countries such as Turkey. Unfortunately, such changes remain undetected in cross-sectional studies. However, panel data covers multiple sequential time periods for the same firms or individuals and allows for analysis of these changing patterns. It is well known that income is one of the most important determinants of socioeconomic status and health. This study aimed to determine predictive factors of income using three different panel datasets. METHODS:  Data from the Income and Living Conditions Survey in Turkey between 2008 and 2011 were used. A Random Effects and Expectation Maximization (RE-EM) tree algorithm was also used in the analysis. RESULTS:  We observed the correlation of marital status, education, labor status, and self-rated health as indicators of income. A decision tree graph shows that education was the most influential predictor of income between 2008 and 2011, and between 2009 and 2011. Conversely, current economic status was an effective predictor variable of income between 2008 and 2009. Health, however, was observed as a non-effective predictor of income in the study years. To compare the RE-EM model’s performance on three datasets, autocorrelation test results and error rates were examined. Results indicate that no autocorrelation was detected using three different datasets (p˃0.05). According to the estimated variance of error, longer periods of data (2008–2011) have lower error rates (0.0385) than shorter periods (0.0389 from 2009–2011 and 0.0290 from 2010–2011, respectively). It is possible that longer data periods lead to lower error rates. CONCLUSIONS:  Experimental results highlight that education outperforms socio-economic status and health as an indicator in the prediction of personal income.