Importance of exploratory data analysis of dummy variables, logit/probit using eviews

By IMPRI Team  
IMPRI Generation Alpha Data Centre (GenAlphaDC) along with IMPRI Impact and Policy Research Institute, New Delhi conducted a Two-Week Immersive Online Hands-On Certificate Training Course on Exploratory Data Analysis with Categorical Variables Regression Models Dummy Variables and Logit/Probit using EViews, on December 10 and 17, 2022. The expert trainer for the course was Professor Nilanjan Banik, Professor at Mahindra University. He is a Visiting Consultant at IMPRI and an Academic Consultant with Geneva Network, United Kingdom, and a Senior Consultant with Hankuk University of Foreign Studies, South Korea.
The convenors for the event were Prof Vibhuti Patel, Visiting Professor at IMPRI and a Former Professor, Tata Institute of Social Sciences (TISS), Mumbai; Dr Soumyadip Chattopadhyay, Associate Professor, Economics, Visva-Bharati, Santiniketan and a Visiting Senior Fellow, IMPRI; Dr Arjun Kumar, Director, IMPRI. The training course had participants from the field of data and policy– including students, professionals, researchers, and many others.

Day 1 | December 10, 2022

The session began by going through the basics of Regression. The first question he pondered upon is what the meaning of a “dummy” is. He stated that in essence, it means a replica. Here, in a regression model, if X is a dummy variable, it means that it is a qualitative variable. He began by laying down some assumptions about the dependent and the independent variables. First, X and the Error Term (e) are not related, if related, there will be a problem of endogeneity. X is not quantitative if it is a dummy variable. He explained how we can constitute various qualitative traits in a dummy variable such as gender, and ethnicity among others in a regression model. It tries to capture the impact of any variable that is qualitative in nature.
Second, he mentioned that a dummy variable can capture any break or shift in data. He used the example of the Indian economic reforms of 1991, which was a breakpoint in terms of per capita GDP levels. After 1991 there was a big jump in GDP growth. In other words, there was a structural break. Dummy variables can capture such structural breaks. Thirdly, he mentioned that dummy variables can also be used to de-seasonalize the data. Using Excel, he showed how to incorporate dummy variables in a regression model and how dropping a dummy variable is important in order to avoid a Dummy Trap. He also showed how to de-seasonalize the data, using Excel. After de-seasonalizing the graph turned out to be more stable than before.
After explaining over Excel, he showcased the same data set on EViews. He selected the variables such as sales figures, trends, etc. He then created the dummy variables out of the four quarters. He ran the regression without the dummy first. Then he showcased a data set where he introduced a dummy and ran the regression. The data set used was US Trends in Gross Personal Income and Gross Personal Savings from 1959 to 2007. The dummy variables reflected the recession points from 1981 to 1984. The regression diagram consequently showed the breaking point due to the recession of 1981. The session ended for the day with this, after which, Professor Banik went on to take questions and clear doubts of the trainees. The next class was saved to learn Logit and Probit Models.

Day 2 | December 17, 2022

The second day of the session conducted by Professor Nilanjan Banik, titled, “Exploratory Data Analysis with Categorical Variables Regression Models: Dummy Variables and Logit/Probit using EViews” was devoted to the concepts of Logit and Probit. Professor Banik started by explaining the basic equation of a regression model, and the components within it. Here, the motive was to explain the concept of dummy variables, and the Probit/Logit model, when the variables X and Y are qualitative respectively. Then with an example of hourly wage rates, he showed how to interpret dummy variables for various categorical variables.
After explaining dummy variables, he followed it up by talking about Logit functions. He mentioned that in Logit functions, the dependent variable, Y, takes values of 1 or 0. The Logit or Probit model describes the odds of an individual meeting the outcome variable, given a certain trait or characteristic. He mentioned the importance of LR tests in Probit models. The Logit/Probit models primarily deal with the dependent variable (Y). He showed that the Y variable takes values between negative infinity to positive infinity. He proved this by showing the method to derive the value of Y using Probability. Since the P value will be between the value of 0 and 1, the Y value will take the value of negative Infinity to positive infinity.
After delving into theory, he started a practical lesson on the above discussions with the help of a data set on EViews. First, he showed how to introduce dummy variables on a set of observations. Then, Professor Banik went on to show how to interpret Logit functions on EViews. For this, he again used the previous US data on Savings and Income to show the recession point. Using other data he showed how smoking is affected by age, income, and education. He explained what the P value shows using the formula for the same. He also showed it practically based on the regression model and the results generated from it. Then he took questions from the trainees which he promptly clarified. With this, the two-day training course ended.

---
Acknowledgement: Aaswash Mahanta is a research intern at IMPRI