RES 6551 Module 1 Data Preparation and Screening Report Example

From 241 Started Surveys to 221 Usable Cases: Data Preparation and Screening for a DBA Study of Leadership, Psychological Safety and Voice in Revenue Cycle Teams

Student Name

American College of Education

RES6551: Analyzing the Dissertation Research

Module 1 Assignment

Instructor Name

October 12, 2026

Introduction

My dissertation examines whether transformational leadership by front-line supervisors is associated with employee voice, the willingness to speak up with suggestions and concerns, among revenue cycle staff in hospitals, and whether team psychological safety carries part of that association. Survey data were collected over five weeks from employees of three health systems in North Carolina and Virginia. Before any research question can be answered, the raw file has to become an analysis file that follows the rules set in Chapter 3. This paper reports that preparation: which cases were removed and why, how missing values were handled, how scales were scored and how the data were screened for outliers and distribution problems.

The Raw File

The survey platform recorded 241 started surveys from employees who passed the eligibility screen, meaning they worked in patient access, coding, billing or denials and follow-up and had reported to their current supervisor for at least three months. Each record contained 20 scale items: seven for transformational leadership from the Global Transformational Leadership scale (Carless et al., 2000), seven for team psychological safety (Edmondson, 1999) and six for voice behavior (Van Dyne & LePine, 1998), along with tenure, work location, role, health system and two attention-check items. The raw export was saved as a read-only file, and every later step was written in an analysis script so that the path from raw to final data can be repeated.

Exclusions

Three preset rules were applied in the order Chapter 3 specified, and the number removed at each step was recorded. First, 9 respondents who answered fewer than 80% of the scale items were removed as incomplete; most stopped within the first block. Second, 8 respondents who failed both instructed-response items were removed. Third, 3 respondents gave the same answer to all 20 scale items, including the reverse-worded psychological safety items, a pattern that cannot reflect genuine agreement with opposite statements, and were removed as straight-line responders. The final analysis sample is 221 cases, 91.7% of those who started, which exceeds the 158 required by the a priori power analysis. The three exclusion counts and the final sample will appear in a participant flow figure in Chapter 4.

Completion Times

Completion time was reviewed but, as planned, not used as an automatic exclusion. The median time for retained cases was 9.4 minutes. Six cases finished in less than one third of that, about three minutes. Each was examined individually: all six had passed both attention checks, none showed straight-lining and their item responses varied in ways consistent with the reverse-worded items. Several of these respondents worked in coding, where staff complete structured forms quickly every day, which offers a plausible reason for speed. All six were kept. Because the decision involved judgment, the main analyses in Module 3 will also be run without these six cases, and any difference will be reported.

Missing Values

Among the retained cases, 17 of 4,420 item responses were missing, or 0.38%. No case was missing more than one item on any scale, and no single item was missing more than three times. A test of whether values were missing completely at random (Little, 1988) was not significant, chi-square(52) = 58.31, p = .26, giving no evidence of a systematic pattern. Because missingness was rare and appeared random, the rule set in Chapter 3 applied: each scale score was computed as the mean of the items a respondent answered, provided at least 80% of the scale's items were present. Multiple imputation, the planned fallback, was not needed.

Scoring

Items 1, 3 and 5 of the psychological safety scale are worded so that agreement indicates less safety, and they were reverse-scored by subtracting each response from 8 before averaging. Each scale score is the mean of its items, so transformational leadership ranges from 1 to 5 and psychological safety and voice from 1 to 7, with higher scores meaning more of the construct. Internal consistency was high for all three: Cronbach's alpha was .92 for transformational leadership, .92 for psychological safety and .91 for voice, each well above the .70 minimum set in Chapter 3. Tenure was recorded in years and work location was coded 1 for remote or hybrid and 0 for on site; 105 of the 221 respondents, 47.5%, worked remotely or hybrid.

Outliers

Univariate outliers were defined as scale scores more than 3.29 standard deviations from the mean, the cutoff for p < .001. None were found; the most extreme score, a voice score of 1.33, was about 3.0 standard deviations below the mean. Multivariate outliers on the three scale scores were checked with Mahalanobis distance against the chi-square critical value of 16.27 for three degrees of freedom at p = .001. The largest distance was 12.34, so no case was flagged. Tenure was right-skewed, as expected for a workforce with many recent hires, with a mean of 4.95 years, a standard deviation of 3.64 and a median of 3.6; it is used only as a control, and its skew does not violate any assumption of the planned regressions, which concern residuals rather than predictors.

Distributions

Skewness and kurtosis for the three scale scores fell well within the range of plus or minus 1 that is commonly treated as acceptable for regression with a sample of this size. Transformational leadership showed skewness of -0.24 and kurtosis of -0.57; psychological safety, -0.37 and -0.53; voice, -0.28 and -0.14. All three were mildly negatively skewed, meaning more respondents rated their supervisor, their team and their own voice above the scale midpoint than below it. Histograms and normal Q-Q plots showed no gaps, ceilings or floors serious enough to distort estimates. Assumptions that concern regression residuals, such as homoscedasticity and normality of errors, will be checked in the module that runs the models.

Collinearity Preview

Because the mediation model includes transformational leadership and psychological safety together, their overlap was checked before modeling. Their correlation was .59, substantial but not close to the levels that make coefficients unstable. Variance inflation factors in the full model were 1.54 for leadership, 1.55 for psychological safety, 1.02 for tenure and 1.00 for work location, all far below the threshold of 5 named in Chapter 3.

Decision Log

Every decision described in this report is recorded in a dated log kept with the analysis script: the order of exclusions, the count at each step, the result of the missing data test, the reverse-scoring of three items, the outlier criteria and their results, and the decision to keep six fast responders. The log also records one correction. An early version of the script reverse-scored the wrong psychological safety item, which produced an alpha of .81 instead of .92; checking item-total correlations exposed the error, and it was fixed before any further analysis.

Conclusion

The final analysis file contains 221 cases after three preset exclusions removed 20. Missing values were rare and random and were handled by within-scale means. Three scales were scored, with reverse-scoring where required, and all showed strong reliability. No univariate or multivariate outliers were found, distributions were close to normal and collinearity is low. The data are ready for the descriptive analysis in Module 2.

References

Carless, S. A., Wearing, A. J., & Mann, L. (2000). A short measure of transformational leadership. Journal of Business and Psychology, 14(3), 389-405. https://doi.org/10.1023/A:1022991115523

Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44(2), 350-383. https://doi.org/10.2307/2666999

Little, R. J. A. (1988). A test of missing completely at random for multivariate data with missing values. Journal of the American Statistical Association, 83(404), 1198-1202. https://doi.org/10.1080/01621459.1988.10478722

Van Dyne, L., & LePine, J. A. (1998). Helping and voice extra-role behaviors: Evidence of construct and predictive validity. Academy of Management Journal, 41(1), 108-119. https://doi.org/10.2307/256902