UNDER-EXPENDITURE DETERMINATION OF ELDERLY PEOPLE AND EXTRACTION OF CHARACTERISTIC EXPENDITURE ITEMS

Abstract

There has been no database suitable for disclosing the influence of dementia on economic activities. Therefore, through using related databases, we have aimed to obtain the principles to resolve its impact. In order to develop the method to detect capability deterioration of economic activities for elderly single-person and two-person households at the age of sixty-five or older, we have employed an anonymous data set obtained from the National Survey of Family Income and Expenditure (NSFIE) carried out by Ministry of Internal Affairs and Communications (MIC) in 1994, 1999 and 2004. Annual income is classified into five groups (High, Mid-H, Mid-L, Low-H and Low-L).

We then propose the detection method mentioned above through selection of features, discriminant analysis, and regression analysis. The ratio of expenditure (after designated total compensation) against annual income is defined as O/I-per. In this method, the higher and lower portions of O/I-per are then selected. Among the expenditure items of the chosen data, the items where expenditure gaps are apt to be outstanding between high/low portions of O/I-per are extracted as feature values and utilized for the analysis. Then for the data whose income is greater than expenditure (after designated total compensation), through the method using the predicted values calculated from the approximate formulas where the feature values above are used as variables, we determine if the data are under-expenditure. Next, for each income group, discriminant analysis is performed by setting O/I-per as variable. Among the data divided into two groups by discriminant analysis, the maximum value among the set of lower O/I-per is set as the threshold to determine under-expenditure. Moreover, approximate formulas are obtained by performing multiple regression analysis, by setting expenditure (after designated total compensation) as respondent variable and the designated feature value as explanatory variable. The prediction values of expenditure (after designated total compensation) is calculated through the formulas obtained. The values are compared with the threshold to determine under-expenditure and used to determine if the data are under-expenditure.

As a result of determination, for single-person household, females show better estimation accuracy than males. In addition, for both single males and females, “the case of not distinguishing rent house with own house” outperforms “the case of distinguishing rent house with own house.” On the other hand, for two-person households, “the case of distinguishing rent house with own house” shows slightly better estimation accuracy than “the case of not distinguishing rent house with own house.”