Organization-level risky decision making was assessed using an objective, global financial measure labeled percent of funds taken. As part of implementing the focal legislation, CMHBs could elect to assume varying levels of responsibility for the care of a high risk population. The level of responsibility assumed by CMHBs is reflected by percent of funds taken and could range from 0% (i.e., no responsibility) to 20% (i.e., maximum responsibility) during the study period. While positive outcomes were possible for CMHBs who chose to take greater responsibility (i.e., greater percent of funds taken), the magnitude of potential financial loss to CMHBs also was directly related to percent of funds taken. As a result, percent of funds taken served as our organizational level measure or risky decision making behavior.


The response rate at the CMHB/organization level was 98% (52/53), with an average of 3.6 responses per Board.
The response rate for the top 4 executive positions (e.g., Executive Director, Associate Director) was 56% (100/179). Responses were received from 36 CEOs or 68% of CMHBs. For 94% (50/53) CMHBs, at least one of the four top executives responded. These rates compare favorably with other research in this area and suggest that respondents are representative of TMTs.

Although hypotheses are tested at the organizational level, the individual was the unit of analysis for assessing the internal consistency of the problem framing (opportunity/threat) scale, perceived risk scale, and the risk propensity scale. Cronbach alphas ranged from .85 to .88 suggesting that scales are internally consistency.

Convergence among key informants was assessed with the inter-rater agreement index (IRR; James, Demaree & Wolf, 1984, 1993) because IRR was developed to compute agreement when judges are nested within firm. Values ranged from a high of .92 for the opportunity/threat scale, to a low of .81 for the inertia measure, suggesting good convergence among respondents within CMHB and providing justification for averaging ratings within CMHB..

Organizational-level correlations indicated all 4 perceptual measures were significantly correlated in the expected directions with the objective indicator of risky decision making. Yet, only 2 of 6 possible correlations among the 4 perceptual measures were significant (and these 2 were expected), reducing concerns about shared method variance. However, contrary to predictions, the link from risk propensity to risk perception was not significant.

The full model (Figure 1) and specific linkages (Hypotheses 1-9) were tested using RAMONA, a structural equation modeling program within SYSTAT (Wilkinson & Hill, 1994). Evaluation of the model followed Medsker, Williams and Holahan's (1994) guidelines. The a priori model and a revised model were compared to a null model with no causal links. Four fit indices were examined to allow for comparisons with the null model: the normed and nonnormed fit indices (NFI, NNFI), the comparative fit index (CFI), which is resistant to errors associated with sample size, and the parsimonious model fit index (PFI), which adjusts for degrees of freedom in the target model. Values of .90 and greater indicate adequate model fit for the first 3 indices; .60 and above has been suggested as the guideline for PFI.

The a priori model demonstrated good fit according to 3 of 4 fit indices. The values of NFI (1.07), NNFI (.97) and CFI (1.00) exceeded the target value of .90. Only, the value of PFI (.32) fell below the guideline of .60. Review of the output from the a priori model indicated that all hypothesized direct paths in the model were significant with the exception of the path from risk propensity to perceived risk. This path was removed from the model and a revised model was run. The values of 3 of the fit indices were essentially unchanged in the revised model and continued to suggest good fit ( NFI =.96; NNFI = 1.1; and CFI = 1.00). The value of PFI (.43) increased somewhat, but was still below the guideline of .60. (Our modest sample size deflates this index). The parameter estimates which are significant at the .05 level for the revised model are shown in Figure 1 and are used to test the 9 hypotheses.

Tests of Hypotheses

Five of seven direct-effect hypotheses were supported. Support was found for H1 (r = 0.21, p < .05), H2 (r = .32, p < .05), H4 (r = 0.71, p < .05), H5 (r = -0.68, p < .05), and H8 (r = 0.22, p < .05). Support was not found for H7 and H9. Hypothesis 7 (H7) predicted a negative correlation between risk propensity and risk perception. Neither the bivariate correlation (r = -0.09, ns) nor the path coefficient (r = -0.08) between these variables were significant. Hypothesis 9 (H9) which predicted a negative association between risk perception and risky decision making also was not supported. The zero-order correlation provided preliminary support for this hypothesis (r = -0.30, P < .05) and the association remained significant but the sign changed from the predicted negative association to a positive association (r = 0.24, p < .05) suggesting the existence of a suppressor effect.

Support was found for H3 which predicted that risk propensity partially mediates the effect of inertia on risky decision making as indicated by the product of the path coefficients (i.e., 0.32 X 0.22 = .06) being considerably lower than the direct path coefficient linking inertia to risky decision making (r = .21, p < .05). H6 predicted that risk perception partially mediates the effect of problem framing on risky decision making. Because the product of path coefficients is much smaller and negative (i.e., -0.68 x 0.24 = -14) compared to the direct path from problem framing to risky decision making, we conclude that risk perception acts as a suppressor, explaining unique variance in risky decision making behavior beyond that explained by problem framing.


Research Home - Page 1 | 2 | 3 | 4 | 5< PREV  NEXT >



Copyright 2002 Decision Support Services, Incorporated.
Website related questions, comments and problems can be sent to