Public Administration: Applying Regression to Research and Evaluation

Week 10 Discussion

Applying Regression to Research and Evaluation

Because public and nonprofit administrators operate in a complex world, they often ask research questions that can be answered with multiple explanations. Multiple explanations for an outcome require multiple independent variables. As a reminder, independent variables are those that influence the outcome or dependent variable. For example, consider research designed to explain why some students earn good grades in Applied Research and Evaluation, while others do not. Perhaps one independent variable, such as time spent studying, explains good grades versus poor grades (dependent variable). However, it is more likely that many independent variables are involved. It could be that time spent studying, along with previous experience taking a research methods course, and the amount of time spent at a job, contributes to the explanation. Simple associational research, such as the research you considered in Week 9, overlooks the complexity of multiple variables because it focuses only on one independent variable.

Explanations that involve multiple independent variables are better served by approaches such as OLS regression or binary logistic regression analysis. The choice often depends on the nature of the dependent variable. An interval dependent variable may lead you to choose OLS regression, while a nominal dependent variable may require binary logistic regression. OLS is a simple, yet limited approach to regression. You must assume that variables are related in a linear fashion, and the dependent variable must be somewhat interval in nature. Research does not always present interval-dependent variables. There are situations, however, in which dependent variables are more nominal in nature. Assume that you want to predict the likelihood of an individual leaving the organization in the next six months. The dependent variable in this example would be nominal (either he/she leaves or he/she does not). OLS regression would not work in this situation, so you would need another type of regression, known as binary logistic regression. There is no “perfect” approach to regression analysis. Rather, you must select the “most appropriate” type of analysis, depending on the purpose of your research and the types of variables you are examining.

Review the Learning Resources for this week. Consider how regression analysis (OLS and binary logistic) could apply to your Final Evaluation Design (Final Project) to help answer your research question.

ASSIGNMENT

WRITE an explanation of whether OLS regression or binary logistic regression is more appropriate to evaluate the program, problem, or policy you selected for your Evaluation Design. Explain how you would use the selected analysis, and justify why this type of regression analysis is most appropriate.

(Introduction to Final Project Below)

Evaluation of policies is essential as it assists in determining their effectiveness. The following paper delineates how the quasi-experimental would be used to evaluate the Eric Garner Anti-Chokehold Act. New York City Police Department (NYPD) is one of the most recognized police departments in the US. However, NYPD has been accused of corruption, transgression, and brutality. Quasi-experimental design will be used to evaluate if the policy will be effective in decreasing police brutality in New York City. The act prohibits the police from applying harmful chokeholds against people. Besides, the policy recognizes aggravated strangulation as a new criminal activity. Some protestors and officials have called for the defunding of police departments. Some mayors plan to defund the police departments in their regions. A quasi-experimental design will also be used to evaluate if defunding the police will eradicate police brutality. The most appropriate evaluation design for the described policy would be quasi-experimental. A quasi-experimental design allows researchers to determine the causal impact of a specific program or independent variable on participants without random assignment. Purposive sampling is the most appropriate sampling method for selecting the sample for the evaluation study. This sampling method requires the researcher to utilize his or her judgment when identifying study participants from a specific population. The potential ethical issues associated with quasi-experimental design include bias, internal validity, confidentiality, and informed consent. The anticipated findings after the evaluation will reveal what would be expected if the research was performed. The policy will decrease police brutality in New York City. Besides, defunding NYDP will also reduce brutality in the region.

Readings

· Johnson, G. (2014). Research methods for public administrators (3rd ed.). Armonk, NY: M. E. Sharpe.

o Chapter 15, “Data Analysis: Regression” (pp. 216–229)

· Garson, G. D. (2009, April 4). Logistic regression. Retrieved from · http://faculty.chass.ncsu.edu/garson/PA765/logistic.htm

Media

· Laureate Education (Producer). (2013a). Correlation and introduction to regression. [Multimedia file] Baltimore, MD: Author.

“Correlation and introduction to regression” Transcript (PDF)