The role of human capital in loan officers' decision policies.

Number of small business loans in the previous year. Respondents were asked to self-report the number of last year's loans that were disbursed to (1) micro businesses and (2) small businesses. To capture the emphasis on smaller businesses, we summed these reported numbers.

Experimental Design

In total there are eight attributes used in the small business scenarios in the experiment each with two levels--high and low. Therefore, there are 256 ([2.sup.8]) possible combinations in the profiles provided to the respondents. Assessing that many profiles would be an overwhelming task for the respondents. Therefore, an orthogonal fractional factorial design was used, which reduced the number of attribute combinations to 16, which makes the decision task more manageable (Green & Srinivasan, 1990). In choosing an orthogonal fractional factorial design we followed the general rule of confounding effects of most interest with effects that are unlikely to be significant or, if they are significant, are unlikely to cause much bias in the parameters that are estimated (Louviere, 1988). We chose a design from Hahn and Shapiro (1966) which confounded the main effects and all two way interactions involving "business risk" (the effects of most interest) with other two-way and all higher order interactions (those effects of least interest); and therefore, it is unlikely that nonhypothesized higher order effects biased this study's results (as higher order interactions typically account for minuscule proportions of variance [Louviere, 1988]). Each of the 16 unique profiles was replicated resulting in a total of 32 profiles. We randomly assigned the 32 profiles and the order of attributes within a profile for four versions of the experiment to test for order effects. There was no significant difference across versions (p >. 10) and therefore order effects are unlikely to have influenced the results.

Analysis and Results

Decision Policy

Before modeling the decision policy of the sample of loan officers, we used regression analysis to analyze each loan officer's set of decisions and ascertain his or her decision policy. A majority of the individual models of loan officers' decision policies (90.4%) explained a significant proportion of variance (p < .001) with a mean [R.sup.2] of .83. Further, Pearson R correlations were computed between each loan officer' s assessment of both the original and the 16 replicated profiles as a test--retest of reliability. A percentage of 96.5 of the loan officers were significantly reliable in their responses (p < .05). The mean test-retest correlation for the sample was .77 (which is high relative to Shepherd's [1999] test-retest reliability of .69). This high degree of judgmental reliability provides assurance that the conjoint task was performed consistently.

Although the experiment provides 32 observations per loan officer and therefore 3,648 observations for the sample, there may be autocorrelation because each of the 32 observations is nested within individuals. Hierarchical linear modeling (HLM) accounts for variance among individuals such that the observations within an individual are independent. (2) We therefore relied on HLM in analyzing the results of the experiment. These results are displayed in Table 2.

As shown in Table 2, all main effects, except for comprehensiveness of strategic plan, in the main-effects-only model are positive and statistically significant, suggesting that they were used by loan officers in their decisions on the likelihood of granting a loan. The full model including the interaction terms, also in Table 2, reports that two of the four hypothesized interaction effects were statistically significant. Specifically, the interactions between business risk and financial position and between business risk and collateral were significant, while the interactions between business risk and the business's share of investment and between business risk and related experience were not significant.

We use hierarchical linear modeling to explain variance across loan officers' decision policies depending on their human capital. HLM produces a model where the particular decision criterion is the dependent variable and the independent variables consist of an intercept, education, banking experience, lending experience, and recent experience processing small business loans. Two control variables, gender and age, are also included in the model. The HLM results for the human-capital variables are reported in Table 3. The decision criteria (dependent variables in HLM model) hypothesized to be influenced by the human-capital variables are listed across the top of the table. The variance in the use of the specified decision criterion explained by the intercepts, education, banking experience, lending experience, and recent experience with small-business loans is shown in the respective rows. (3)

Hypothesis 1 stated that loan officers with more human capital place greater emphasis on contingent relationships that mitigate business risk. Surprisingly, of the sixteen potential relationships between human-capital variables and decision contingencies, only one was significant at p < .05. General human capital, or education, significantly explains variance in the use of the contingent relationship between business risk and financial position (coefficient = .522, p < .05), but this finding is in the opposite direction of that hypothesized in hypothesis la. Bank experience and lending experience did not significantly explain variance in the use of decision contingencies, thus, Hypothesis 1b and 1c are not supported. Yet, recent small-business loan experience is marginally significant in explaining variance in the use of the interaction of business risk and the independence of collateral (coefficient = .003; p = .054). This finding provides marginal support for hypothesis 1d.

Hypothesis 2 focuses on the outcome of the decision process, and the results can be determined by examining the intercept coefficients for the specific human-capital variables lending experience and recent experience with small-business loans. As reported in Table 3, the intercepts for these variables were not significant, indicating that these human-capital variables do not significantly influence the probability that a loan officer will favorably assess loan applications by small businesses (holding constant all decision attributes). Hypothesis 2 is not supported.

Hypotheses 3 and 4 offer a different perspective, proposing that the emphasis placed on the importance of related business experience in making a loan decision is dependent on the degree of similarity between the loan officer's human capital and the business owner's human capital. As shown in Table 4, three of the four human capital variables (lending experience, recent experience with small-business loans, and education) significantly explained variance in the emphasis placed on the business owners' specific human capital by loan officers in their decision processes. The coefficient for general bank experience was not significant, so hypothesis 3a is not supported. To analyze the results for hypotheses 3b, 3c, and 4 we must first look at the main effect for related business experience on the sample as a whole. Table 3 shows that the main effect was positive (coefficient = 1.269, p < .001). Therefore, a positive coefficient for lending experience (coefficient = .022, p < .05) implies that loan officers with a higher level of lending experience placed more emphasis on the business owners' related experience than those loan officers with a lower level of lending experience. Thus, hypothesis 3b is supported.

Similarly, a positive coefficient for recent experience with small-business loans (coefficient = .003, p < .001) sustains the argument that loan officers with more specific human capital will value the business owner's specific human capital more than loan officers with less specific human capital, providing support for hypothesis 3c.

In contrast, a negative coefficient for education (coefficient = -.313, p < .01) indicates that loan officers with more education place less emphasis on the borrower's related competence than loan officers with less education while controlling for age. Hypothesis 4 is supported.

Discussion

There are several theoretical and applied contributions of this study. First, we found evidence that there is heterogeneity in the decision policies across loan officers. Prior research on the decision policies of resource providers (venture capitalists and banks) has typically explicitly or implicitly treated these decision makers as a homogenous set of individuals that will ultimately reach a similar conclusion for any given business owner. A major reason for implicitly assuming that loan officers constitute a homogeneous group is that banks develop lending guidelines, which should increase decision homogeneity among lending officers. However, these guidelines can only relate to characteristics of the potential borrower that are known with relative certainty. Because of information asymmetry, many aspects of the small business are uncertain, requiring personal judgment on behalf of the loan officers, and according to our findings, loan officers make differential judgments depending on their level of human capital.