The purpose of this study was to describe relationships among organizational context (characteristics of the environment and technology) and structure (relational coordination) fit and organizational effectiveness (length of stay and severe medication errors) on medical- surgical nursing units. In this chapter, results from the analyses of models to separately test the relationships among contextual-structure fit and patients‟ length of stay and severe medication errors are reported. This chapter begins with a description of the variables that were included in the research models for this study. Following this description, the major study findings are presented. This discussion will start with an evaluation of these data in terms of the assumptions of the underlying statistical procedures that were used and conclude with the results related to the hypotheses as stated in Chapter 3.
Description of Study Variables
Major Study Variables
Table 11 summarized the descriptive statistics for the major study variables. Units in this study were moderately large with an average of 34 beds. Workload was somewhat higher than the national average of 5 patients to 1 nurse (Altman et al., 2005), with an
average for these units of 5.27 patients per nurse with scores clustered around 3.6 to 6.94. In general, support service availability was rated as medium to high as was work complexity. Nurses on these units rated patient acuity as medium to low with only some to about one-half of patients identified as needing frequent and more technologically complex patient care.
133 Table 11
Descriptive Statistics for Study Variables
Study Variables Mean SD Minimum Maximum
Major Study Variables
Unit Size 33.53 11.16 13.00 80.00
Workload 5.27 1.67 2.26 12.00
Support Services Availability 32.36 2.49 23.00 39.17
Patient Acuity 45.57 3.59 34.50 56.67
Work Complexity 26.84 3.50 15.79 37.40
Relational Coordination 226.06 12.67 157.50 262.80
Relational Coordination
Nurse-Physicians & Nurse Pharmacist
51.12 3.52 34.00 59.67
Length of Stay 4.55 1.09 2.23 9.22
Severe Medication Errors 2.00 3.67 0.00 29.00
Control Variables
Hospital Size 345.84 185.22 75.00 1242.00
Teaching Status 0.13 0.25 < 0.01 1.23
Case Mix Index 1.44 0.32 0.89 3.67
RN Experience 138.49 45.38 43.57 322.80
Unit Tenure 74.77 33.05 9.00 199.89
Educational Preparation 36.52 19.36 < 0.01 100.00
134
The quality of relational coordination among all healthcare providers was rated as medium to high on most units. Similarly, the quality of relational coordination among nurses,
physicians, and pharmacists was also rated as medium to high. The average length of stay on these units was 4.55 (SD = 1.09) with a range from 2.32 to 9.22 per 1000 patient days. On average, two medication errors (SD = 3.6) were reported on these units during the final three month data collection period of the ORNA-II study. Despite this low number, however, the frequency of medication errors during these three months was widely dispersed, with medication errors ranging from 0 to 29.
Control Variables
Based on the review of literature, the decision was made to control for three hospital- and four unit-level variables during model testing. At the hospital level, the potential effect of hospital size, teaching status, case mix index were controlled. At the nursing unit level, the potential effect of three RN workgroup characteristics were controlled including average nursing experience, average unit tenure, and percentage of the RN workgroup with a
baccalaureate degree in nursing or higher. In addition, the potential effect of skill mix in terms of the composition of the nursing workgroup on each unit was controlled. A summary of the descriptive statistics for the control variables are provided in Table 11.
Multicollinearity Among Study Variables
Multicollinearity can be defined as the presence of two or more independent variables that are strongly, but not perfectly, correlated (Berry & Feldman, 1985; Fox, 1991).
Although multicollinearity does not violate the assumptions of regression, its presence tends to inflate standard errors of the regression coefficients. In other word, the result of the regression analysis may be biased when variables in the model are strongly correlated
135
Table 12
Bivariate Correlations Among Study Variables
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Major Study Variables
1. Unit size 1.00 2. Workload 0.13 1.00 3. Support Services 0.04 -0.16 1.00 4. Patient Acuity < 0.01 0.02 -0.11 1.00 5. Work Complexity 0.23** 0.16 -0.23** 0.17* 1.00 6. Rel Coord -0.05 -0.06 0.29** 0.04 -0.32** 1.00 7. Rel Coord (MD/PH) -0.10 -0.08 0.26** -0.02 -0.33** 0.83** 1.00 8. Length of Stay -0.11* 0.09 -0.09 0.21** -0.13 0.09 0.01 1.00 9. Medication Errors 0.15** -0.07 0.05 -0.05 -0.03 0.07 0.08 0.06 1.00 Control Variables 10. Hospital Size 0.07 0.03 0.14 0.07 0.04 -0.06 -0.11 0.22** -0.02 1.00 11. Teaching Status -0.08 0.17* -0.12 0.03 0.03 -0.17* -0.31** 0.01 -0.09 0.33* 1.00
12. Case Mix Index -0.03 0.23* -0.12 0.04 0.09 0.04 0.03 -0.03 0.01 0.38** 0.24** 1.00 13. RN Experience 0.11 < -0.01 -0.07 -0.09 -0.05 -0.04 0.03 -0.08 0.04 -0.05 0.02 -0.01 1.00 14. Unit Tenure 0.04 -0.07 .07 -0.11 -0.18* 0.09 0.16 -0.07 0.09 -0.05 -0.03 0.07 0.62** 1.00 15. RN Education 0.01 -0.17* -0.08 0.12 0.06 0.02 -0.11 0.07 -0.14 0.26** 0.30** 0.26** -0.05 -0.04 1.00 16. RN Skill Mix -0.09 -0.56** 0.13 -0.02 -0.20* 0.09 0.10 -0.02 0.04 0.23** 0.26** 0.14 0.03 0.05 0.20** 1.00 Note. Significance levels have been adjusted to account for clustering of units inside the same hospital for correlations involving length of stay (by LMM) and medication errors (by random effect negative binomial models). For all other correlations the degrees of freedom were adjusted: *p < 0.05; **p < 0.01.
136 Table 13
Evaluation of Data for Multicollinearity
Study Variables Length of Stay Medication Errors
Tolerance VIF Tolerance VIF
Major Study Variables
Unit size 0.88 1.13 0.88 1.13
Workload 0.64 1.57 0.63 1.57
Support Services Availability 0.80 1.24 0.82 1.22
Patient Acuity 0.93 1.08 0.94 1.06 Work Complexity 0.76 1.31 0.76 1.31 Relational Coordination 0.80 1.25 Relational Coordination MD/PH 0.75 1.34 Control Variables Hospital Size 0.74 1.35 0.75 1.34 Teaching Status 0.74 1.35 0.69 1.44
Case Mix Index 0.77 1.3 0.77 1.30
RN experience 0.57 1.75 0.58 1.72
Unit Tenure 0.56 1.77 0.81 1.24
Educational Preparation 0.81 1.22 0.81 1.24
137
(Bahn & Massenburg, 2008; Schroeder, Sjoquist, & Stephan, 1991). There are several ways to evaluate for the presence of multicollinearity. First, the magnitude of the zero- order correlations in a bivariate correlation matrix can be examined. Correlations that exceed 0.80 indicate a high potential for multicollinearity (Bahn & Massenburg, 2008; Berry, 1993). In Table 12, bivariate correlations among the study variables are reported. Although some correlations were statistically significant, none exceeded the value of 0.80, suggesting that prima facie evidence of multicollinearity was not present in these data. Second, multicollinearity can be evaluated by inspecting the values for the variance inflation factor (VIF), defined as 1/(1-R²), and tolerance, defined as 1-R². VIF values are one when there is no multicollinearity and increase as multicollinearity increases.
Tolerance values of one indicate no multicollinearity and decrease in the presence of multicollinearity. Table 13 summarizes the results of the regression analyzes that were used to calculate VIF and tolerance. VIF values ranged from 1.13 to 1.67 and tolerance values ranged from 0.61 to 0.93, suggesting limited multicollinearity among the study variables.
Analysis Using Length of Stay as the Dependent Variable
Random-effects mixed models were used to describe the relationships among context-structure fit and length of stay. Evaluation of model fit will be discussed prior to reporting the results of hypothesis testing. As reported in Chapter 4, unit tenure was excluded as a control variable in this analysis because no literature was found to suggest a relationship between unit tenure and length of stay.
Model Fit
138
139
140
which the interaction terms for contextual-structural fit were excluded are shown in Figure 8. Plots for the final model in which interaction terms were included are shown in Figure 9. In both figures, the residual histogram with overlaid normal density are
graphically displayed in the upper right corner. Values for Akaike‟s information criterion (AIC), the Bayesian information criterion (BIC), and the finite-population corrected AIC (AICC) are reported in the lower right corner. The information in these figures suggests that these data approximated a normal distribution, met the distributional assumptions of mixed models, and were appropriately matched to the chosen variance function.
According to Littell, Milliken, Stroup, Wolfinger, and Schabenberger (2006), smaller values for the AIC, BIC, and AICC are indicative of a better fitting model. As shown in Figures 8 and 9, smaller values for these indices were found for the main model,
suggesting that this model provided a better fit to the data than did the interaction model. However, the log likelihood ratio test indicated that the main and interaction models did not significantly differ (df = 5, chi-square= 7.9, p = 0.16).
Hypothesis Testing
Results from the analysis of the random-effects mixed models for length of stay as the dependent variable are reported in Table 14. Results from the analysis of the main model were used to address the first hypothesis. Results from the analysis of the
interaction model were used to answer all remaining hypotheses. In this analysis, the value for relational coordination was based on the sum of the ratings for quality of relational coordination between nurses and nine other healthcare providers.
Hypothesis 1. Hypothesis 1 stated that nursing units with higher quality relational