A regression analysis is an approach to
sorting variables out using a mathematical approach to see which variables do
or do not have an impact (Gallo, Davenport, Kim, 2017). A regression analysis
helps us to answer which factors from these variables are going to matter most
and which will matter least. It measures interaction between the factors of the
variables. The most important measurement benefitted from a regression analysis
would be the certainty of all these factors (Gallo, Davenport, Kim, 2017).
The first part in
conducting a regression analysis is to ask the right questions and design the
study that will help answer those questions (“The 13 Steps,” n.d.). Once the
research questions are clearly defined the study needs to be designed to obtain
the answers for those very questions. The design of the study may be one of the
hardest parts since it involves randomization and sampling to find the right
study to answer the right questions (“The 13 Steps,” n.d.).
After the study is
designed the data needs to be gathered on the variables in question. The
regression analysis will consist of at least two variables, a dependent
variable and an independent variable. The dependent variable is the part of the
analysis that is predicted beforehand, the one that needs further understanding
(Gallo, Davenport, Kim). The independent variable is the factor that is assumed
to have impact on the dependent variable (Gallo, Davenport, Kim, 2017). The
data is then gathered from these variables. This data can be based on nominal, ordinal,
or interval measurements (“The 13 Steps,” n.d.).
The second part of
the regression analysis is preparing and exploring the data. When the data that
will be measured is decided upon then it will be time to collect and enter the
obtained data. Depending on the model used for the analysis, the next steps may
vary when it comes to entering the data in. An analysis plan should be created
prior to entering in the data, and that plan will determine which model is used
for entering the variables (“The 13 Steps,” n.d.). The model should then be ran
according to the analysis plan. While this model may not be the final model, it
should closely resemble the right kind of model to best fit the variables,
design, and the research question (“The 13 Steps,” n.d.).
The third part of the
regression analysis is to edit any predictors and make sure the model used is
the best fit for the analysis. There are a variety of stepwise approaches that
can be used to best determine predictors for models set up for predicting
purposes. If the model is set up to answer theoretical research questions then
the model may need to be refined.
If a model needs to
be refined it can be done in a number ways. Interactions and quadratic can be
tested and if need-be, dropped, to explore non-linearity types of models (“The
13 Steps,” n.d.). Control variables that are not obviously significant can also
be dropped (“The 13 Steps,” n.d.). Hierarchical modeling can be produced so
that the results from the predictors can be seen by themselves or in groups(“The
13 Steps,” n.d.). Over-dispersion should be checked and random effects can be
tested (“The 13 Steps,” n.d.).
After refining the
models, data issues need to be checked for and resolved. This will be checking
for data issues that are within the models, but are not classified as data
assumptions. Data issues can include: multicollinearity, outliers and
influential points, data that is missing, truncation and censoring (“The 13
Steps,” n.d.). None of these issues will show up until the selected variables
have been chosen and inputted into the model.
Finally, the results
are interpreted. The results are then used by companies to make smarter
business decisions (Gallo, Davenport, Kim, 2017). The results may be used for
finding ways to increase sales (Gallo, Davenport, Kim, 2017). Employee
retention or recruitment can also benefit from regression analysis (Gallo,
Davenport, Kim, 2017). Generally, businesses use it to gain explanations for
specific occurrences that they may not be able to understand or to predict things
concerning future outcomes for their business (Gallo, Davenport, Kim, 2017).