Selection of models in contingency tables: A reexamination

Effects of Small Business Accounting Bases and Accountant Service Levels on Loan Officer Decisions

“…Survey results are contradictory about loan officers' perceptions of confidence in reports with a service level less than an audit and the ability of the report to permit an assessment of risk (Edmonds et al, 1981;and Diamond et al, 1981). Studies of loan officers' decisions involving audited and reviewed financial reports found that loan officers are more confident with audits but that there is no significant effect of the service level on the loan decision or interest rate charged and the maximum line of credit offered (Johnson et al, 1983;and Miller, 1985).…”

“…Because responses to the loan approval question are binary, they do not represent a normal distribution and are tested using a logit approach. This approach involves building log-linear models that control for (include) the main effects, service level and accounting basis, and their interaction and then testing the significance of the models (Hinkle and McLaughlin, 1985). As shown in Table 1, the values of@ are substantially greater than any reasonable significance level so that hypotheses l a , 2a, and 3a cannot be rejected.…”

“…Yet, surveys indicate that loan officers' perceptions of their needs differ from accountants' perceptions of loan officers' needs (Edmonds et al, 1981;and Nair and Rittenberg, 1983). Additionally, the research to date has focused on loan officers' opinions rather than their perceptions and actions, with limited exceptions (Johnson, et al, 1983;and Miller, 1985). These exceptions, however, considered only the service level issue.…”

Using log‐linear models in higher education research

“…There are two types of log-linear models, association models and togit models. The purpose of association models is to determine the independence or association between variables (Hinkle and McLaughlin, 1984). Kennedy (1983) refers to these models as "symmetrical" because the research question concerns only the presence or absence of an association between two variables, say A and B.…”

“…For the logit model of Table 2, the post hoc procedures would be standardized residuals (Hinkle, Wiersma, and Jurs, 1988) or contrasts analogous to Tukey or Scheffe methods (Marascuilo and McSweeney, 1977). To illustrate the use of hierarchical log-linear models for multiple categorical variables, consider the data discussed by Hinkle and McLaughlin (1984) and found in Table 3. In this example, the researcher was interested in the dropout rates for college students cross classified on the basis of race and gender.…”