The costs of conducting a statistical survey are often a decisive factor in selecting a survey design, but at the same time we do not have many elaborated instruments for dealing with them, at least compared to errors in the survey.
The costs of the survey limit the range of survey designs and the sample size, therefore they also contribute significantly to the risk of survey errors which are elaborated in great detail in the statistical profession. The Master's thesis discusses the problem of linking costs and errors on the example of the survey Tourism Travels of Domestic Population, which is conducted by the Statistical Office of the Republic of Slovenia (SURS).
For the reference period of the 3rd quarter of 2021, SURS collected data on tourist trips from two independent samples, in addition to the regular web and telephone mixed-mode data collection, it also collected data from an independent sample, through computer assisted personal interviewing. The Master's thesis analyses the differences in the accuracy of survey estimates and presents the costs of conducting both surveys.
I calculated which sample design has the lowest relative mean squared error (rMSE), a common measure of estimation accuracy, within an assumed fixed cost limits.
Since each statistic has a different rMSE, I analyzed rMSE on a set of five key target variables.
Since I do not know the true population value for the key statistics, we calculated the relative mean squared error using simulations.
In addition, I calculated the rMSE for three statistics for which there is a comparable registry population value.
The aim of this Master's thesis is to determine the differences in the accuracy of estimates collected through mixed-mode, web and telephone, survey and independent face-to-face survey in the statistical survey Tourism Travels of Domestic Population, to create a model for estimating costs of compared sample designs and to find the optimal sample design, according to the costs and accuracy of the estimates.
By comparing the estimates, I find that the difference between the key statistics of the two survey designs is statistically significant in four of the five considered key statistics.
In the case of three statistics, where we have an external population value from the register, a mixed-mode design is more accurate assessment in all of them.
By evaluating the costs, I find that the total costs of conducting the mixed-mode survey per completed questionnaire unit were lower in mixed-mode design than in face-to-face design.
With comparison of the results of all eight considered statistics, I evalauted the mixed-mode design as the more appropriate.