Browsing by Author "Junes, Tara"

In the analyses performed for the Finnish subsample of European Community Household Panel (ECHP) it was noticed that unit nonresponse bias at the beginning of the panel faded away within time in the case of certain income variables. This kind of a research result would have substantial effects on the estimation based on panel studies and on the research arrangements. In practice the strengthening of the research result would mean that the estimates of panel studies would become unbiased in time. In that case it would be more supporting to use long panels instead of short-terms. The objective of this Thesis is to investigate the fade-away hypothesis with a different dataset and to show that the existence of the fade-away effect is not so straightforward as the results received earlier from the ECHP study pointed out. In this Thesis the main attention is given to the successor of the Finnish ECHP namely the Finnish subsample of EU Statistics of Income and Living Conditions (EU-SILC). EU-SILC is a panel study with rotational sampling design with a rotation period of four years. In this Thesis one rotation group whose register incomes come from the years 2005--2008 is selected for the analyses. The main analysis variable of this Thesis is disposable household equivalised income which is the total disposable household income adjusted by the household composition. To analyse the effects of unit nonresponse the dataset is divided into three groups with a different response profile. The transitions of the members of groups between income quintiles are examined with the help of empirical distributions and furthermore, the transitions are modelled with Markov chains. In this Thesis it is noticed that in the initial wave of the panel there is only a small amount of unit nonresponse bias. Here the initial wave refers to the year 2005. Within time the income distribution of the respondents into the quintiles computed from the actual sample becomes more biased which is caused by the panel attrition. Furthermore, it is shown with the modelling based on the Markov chains, that the unit nonresponse bias will increase after the four years analyse period. So, the estimates that have been computed from the analysed panel become unbiased in time which questions strongly the presented fade-away hypothesis.