RES 6521 Module 4 Data Collection and Analysis Procedures Example

From Survey Link to Simple Slopes: Data Collection and Analysis Procedures for a Moderated Regression Study of Accountant Turnover Intention

Student Name

American College of Education

RES6521: Research Methodology

Module 4 Assignment

Instructor Name

November 2, 2026

Introduction

Earlier modules fixed the design, the sample and the measures. This paper describes what will happen from the moment the survey opens to the moment results are ready to report. It covers the collection timeline, how data will be stored and prepared, how incomplete and careless responses will be handled, which statistical assumptions will be checked and how, the regression models that answer each research question, how an interaction will be interpreted if one appears and how shared method variance will be estimated. Every rule is set now, before any data exist, so that decisions cannot be shaped by the results.

Data Collection

The survey will run on a paid Qualtrics account held for the study, with anonymous links and IP address collection turned off. After IRB approval and the pilot, the link will be released through the three state CPA society channels and two online groups on the same Monday, timed for late May so that it falls well after the spring busy season and before summer vacations. Reminders go out after one and three weeks. The first screen presents the consent information; respondents who agree move to the eligibility questions, and those who qualify continue to the main survey. The survey stays open for six weeks, longer if the 195-start target has not been reached by then. A daily log will record started and completed counts by recruitment channel, using a hidden field attached to each channel's link.

Data Handling and Storage

Responses will be exported in a single file after collection closes, stored on an encrypted drive and backed up to an encrypted cloud service that requires two-factor sign-in. No names, firm names or email addresses are in the survey file; drawing entries sit in a separate form account and will be deleted once prizes are sent. Analysis will be done in R, with every step written in a script so that the full sequence, from raw export to final tables, can be rerun. Raw data will be kept unchanged, and all cleaning will happen in the script.

Screening and Missing Data

Screening follows the rules set in Module 2 and is applied in a fixed order: ineligible respondents, then responses failing both instructed-response items, then speeds faster than a third of the median completion time, which are reviewed rather than removed automatically. The count removed at each step will be reported. For the remaining cases, item-level missing data will be summarized, and the test Little (1988) proposed will be used to examine whether missingness appears completely random. If fewer than 5% of values are missing on the scales, scale scores will be computed when at least 80% of a scale's items are answered. If more is missing, or the pattern is not random, multiple imputation with 20 imputed data sets will be used, and results will be compared with complete-case results.

Preliminary Analyses

Before testing hypotheses, the analysis will report descriptive statistics for all variables, the sample's makeup by state, firm size, service line and title, and correlations among the study variables. Internal consistency of the two scales will be estimated, and a confirmatory factor analysis will test whether support and turnover intention items load on separate factors, as planned in Module 3. Regression assumptions will then be checked on the final model: linearity through plots of residuals against predicted values, independence of errors, homoscedasticity through the same plots and a formal test, normality of residuals through a Q-Q plot, and multicollinearity through variance inflation factors, with values above 5 treated as a concern. Influential cases will be identified with Cook's distance; they will not be removed, but results will be reported with and without them if conclusions change.

Testing the Research Questions

Both questions are answered with one hierarchical regression in three steps. Step 1 enters the four controls: tenure, age, busy-season hours and firm size band. Step 2 adds remote work intensity and perceived organizational support, both mean-centered so that their coefficients describe effects at average values of the other variable (Aiken & West, 1991). The first research question is answered by the coefficient for remote work intensity in Step 2 and the change in explained variance from Step 1. Step 3 adds the product of the two centered variables. The second question is answered by the interaction coefficient, its 95% confidence interval and the change in R-squared from Step 2. The same model will be estimated with the PROCESS macro's simple moderation model, which provides bootstrapped confidence intervals as a check (Hayes, 2022).

Interpreting an Interaction

If the interaction is statistically significant, a coefficient alone will not show what it means for accountants, so it will be probed in two ways. Simple slopes will estimate the relationship between remote work intensity and turnover intention at low, average and high perceived support, defined as one standard deviation below the mean, the mean and one standard deviation above. The Johnson-Neyman technique will then identify the range of support scores over which the relationship is statistically significant, which avoids relying on arbitrary cut points (Hayes, 2022). A plot of predicted turnover intention across the range of remote work, drawn separately for low and high support, will be the main way the result is communicated to firm leaders. If the interaction is not significant, the main-effects model from Step 2 will be interpreted and the nonsignificant interaction reported with its confidence interval.

Estimating Method Variance

The marker item placed in the survey to be theoretically unrelated to the study variables will be used to estimate common method variance. In the approach of Lindell and Whitney (2001), the smallest correlation between the marker and the substantive variables will serve as an estimate of method-related correlation, and the key correlations will be adjusted for it. If adjusted correlations remain meaningful, method variance is unlikely to explain the results. This check supplements the procedural remedies built into the survey layout rather than replacing them (Podsakoff et al., 2003).

Sensitivity Checks

Two sensitivity analyses will show whether the conclusions hinge on particular analytic choices. First, the main model will be re-estimated using the busy-season remote work figure instead of the non-busy-season one, since accountants' arrangements may shift when deadlines arrive. Second, because the turnover intention scores may be skewed toward the low end, the model will be rerun with heteroscedasticity-consistent standard errors. If either check changes a conclusion, both versions will be reported and the difference discussed in Chapter 5. Neither check will be used to choose the more favorable result.

Reporting

Chapter 4 will present results in the order of the research questions. Tables will include descriptive statistics and correlations, the three regression steps with unstandardized and standardized coefficients, standard errors, confidence intervals, R-squared and its changes and the simple slopes. Significance will be judged at an alpha of .05, but effect sizes and intervals will be reported for every test so that readers can judge practical importance. Any departure from this plan will be described and justified in Chapter 4.

Conclusion

The procedures run from a six-week online collection with channel tracking, through scripted cleaning and preset screening rules, to a three-step regression that answers both research questions. An interaction, if present, will be probed with simple slopes and a Johnson-Neyman range and shown in a plot; method variance will be estimated with a marker item. Module 5 adds ethics and limitations and assembles the complete Chapter 3.

References

Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. SAGE Publications.

Hayes, A. F. (2022). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (3rd ed.). Guilford Press.

Lindell, M. K., & Whitney, D. J. (2001). Accounting for common method variance in cross-sectional research designs. Journal of Applied Psychology, 86(1), 114-121. https://doi.org/10.1037/0021-9010.86.1.114

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

Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879-903. https://doi.org/10.1037/0021-9010.88.5.879