Variable selection for classification and forecasting of the family firm's socioemotional wealth

Socioemotional wealth (SEW) refers to those family-centered goals that are likely to have a major influence on the strategic decision-making process and performance of family firms. Many studies have used indirect indicators related to family involvement in ownership and management to measure SEW; meanwhile, others have developed scales to directly measure the level and importance of SEW in family firms. Limitations of both indirect and direct measures of SEW lead empirical research on SEW to be under threat. In the current study, we use random forests to identify the important indicators related to financial and economic decisions, as well as family-related measures, for explaining the family firms' SEW and to design a good prediction model using the smallest set of nonredundant indicators. Our results show that the model that exhibits the minimum out-of-bag sample (OOB) error rate includes variables that refer to the presence of family members in the firm's management positions, long-term nonfinancial debt, personnel expenditures, long-term financial investments, short-term financial debt, average storage period, and accounts receivables. For prediction, the model with a reasonably low estimated classification error includes only three variables, which refer to the presence of family members in the firm's management positions, long-term nonfinancial debt, and accounts receivables.