Att minska variabelbortfallsfelet - En fallstudie av Strukturundersökningen 2020

Sammanfattning: The ideal survey is a survey without errors. This essay seeks possible improvements to diminish the item non-response error in order to minimize the total error. It’s done by investigating and analyzing the results from the Census of Agriculture 2020 made by the Swedish Board of Agri-culture. Normally an evaluation of a survey and its errors is useless since the survey is already conducted and the error can’t be taken away. To evaluate the results from the Census of Agri-culture is however valuable since the survey is recurrent and conducted regularly. The results from this essay can therefore be useful for future purposes. The results are also in some extent possible to generalize to other surveys. This essay analyzes the part of the questionnaire containing questions about employment. It was chosen since it’s a part of the questionnaire with a high frequency of item non-response. To investigate the data the questions where the response rate between the different survey modes (paper and web) were the highest were analyzed. Chi-square-tests were conducted to see whether the differences were significant or just caused by randomness. The data analysis is supplemented with a graphical analysis of the questionnaire. The conclusions are based upon these analyzes and the results show that some adjustments could be made in the layout of the web survey. Another conclusion is that the respondents answer more questions in the web sur-vey and for future censuses more people could be directed towards the web questionnaire to diminish the item non-response. Possible sources of error are also investigated, such as the res-pondents’ characteristics and their tendency to answer.