Assessment of the impact of design errors and respondent failures on household survey results

Assessment of the influence of uneven representativeness of households with different income levels in the sample on the estimation of the parameters of the lognormal distribution of per capita monetary income based on the data from the survey “Selective Observation of Population Income and Participation in Social Programs” (SNPI). These surveys became the basis for analyzing the distribution of household incomes of different socio-demographic types. At the same time, Goskomstat uses the hypothesis of the lognormality of the distribution of per capita cash income, the parameters of which are estimated on the basis of sample data and balance sheet statistics. The problem of estimating parameters is complicated by errors in the design of the sample and various kinds of refusals of respondents to participate in the survey. An incorrect estimate of the average values of per capita income in extreme deciles leads to an underestimation of the fund ratio, which is one of the indicators of income inequality of the population. The article proposes a method for assessing distribution parameters using a criterion different from that adopted in statistics. A model for assessing missing data that was lost due to design errors and survey refusals was proposed. The critical boundaries are constructed, above which there are practically no observations in the sample. Estimates of the per capita income of households not included in the sample are calculated, the addition of which brings the average sample per capita income to the official level obtained from the balance of income and expenditures of the population. The article presents full calculations for a sample of households of the population of the Russian Federation for 2016 and the final results for the rest of the years.