The regression discontinuity design involves the two group evaluation model that is the posttest and pretest. For regression discontinuity model, the statistical model is used to represent the value of pre test and coded values of dummy variables in order to operate the program. As the regression discontinuity analysis is the examination of the regression discontinuity design, so in order to understand the model few of the assumptions are important to portray such model. The assumptions for the model are;
The Cutoff Criterion
The cutoff criterion should be pursuing without any exemption. A biased threat arouses when in the program there appears the wrong cutoff values. Mismanagement in case of the cutoff is termed as fuzzy RD design which explains the difficulties in that are beyond the scope.
The Pre Post Distribution
the pre post distribution is explained as the polynomial function and if the pre post relation appears to be true in case of logarithm, exponential or any other then the program is supposed to be biased. For instance if the data is stated as polynomial distribution more preferable then the model is more likely to be problematic to understand and if the relationship that exists is not a polynomial then high order polynomial sufficiently notice for no matter what function exists. [sky]
Comparison Group Pretest Variance
A model contains enough number of pretest values in order to attain the true relationship.
Implementation of the program is very significant in a view that it is delivered appropriately to the recipients.
Steps in the Analysis
The regression continuity comprises of the following steps.
Transform the Pretest
The analysis starts from the subtraction of the cutoff value from the pretest value. It can be represented as:
Mean (Xi) = Xi-Xc
This equation is evaluated to become the cutoff value similar to the intercept.
Visually Examine Relationship
In a pre-post relationship, there are two things to consider. First is to visualize the discontinuity relationship at the cutoff point. The discontinuity change is examined vertically or a change in a slope or in both the cases. Second is the determination of the bivariate relationship as in the degree of polynomial which is indicated at the slope of distribution.
Specify Higher Order Terms and Interactions
The rule of thumb is important in this step that is the number of flexion indicates the orders of the polynomial. In such a case, the bivariate distribution is linear that determines no flexion points.
Estimate Initial Model
In this step, the analysis is ready to start. For the accomplishment of the model any multiple regressions can be used. If in the previous step, the polynomial function is determined properly then the results appeared to be unbiased. The interaction step is also analyzed and the direction is based on the sign of coefficient and the direction to the scale. [linkunit]
Refining the Model
The step that is refining the model is somehow tricky in order to have the unbiased results. This step is most generally based on the examination of previous step. The final position of the model is significant and with higher efficiency. In the whole process many of the terms might be dropped because to avoid biased results.
• Gellman, A. and Hill, J., Data Analysis Using Regression and Multivarial Model. Cambridge University Press.
• Lemieux, T. (2008) “Journal of Econometrics”. Regression discontinuity designs: A guide to practice. Volume 142(2), Pp: 615–635.