Analytics and Empirical Modeling

 MGT 6503 Exam II Study Guide

Topic: Analytics and Empirical Modeling

Read: Class Handout (Part I) HAN, CHRIS, CARSON

1.  What are the two approaches that you can use to estimate the payoff from pursuing an MBA? What are the advantages and disadvantages of the two approaches?

1. Cross-sectional Analysis (one person, one data point, control variables)

1. Advantage(s)

1. Relatively inexpensive and takes up less time to conduct

1. Disadvantage(s)

1. Difficult to determine temporal relationship (lacking time elements)

1. Panel Data Analysis (one person, data point for each period. time varying control variables only)

1. Advantage(s)

1. More accurate inference of model parameters

1. Disadvantage(s)

1. More difficult/expensive to collect data

2.  What are the two purposes of building analytical models? How will your focus on interpreting the results change based on your purpose?

The two purposes of building analytical models are to 1) discover relationships and 2) predict and forecast outcomes/effects. The two purposes will vary due to inclusivity of certain variables. Discovering relationships will generally involve more variables to see where relationships lie in order to discover causes. Predicting/forecasting focuses more on significant variables to predict future values.

3.  Understand how to interpret the results of cross-sectional analysis shown on Page 7.

Constant coefficient of 16653.33 means that the regular wage of a female worker without an MBA degree with no education and no experience equals to $16,653.33. Education, Experience, Male, iqtest, mba, and height all came out to be significant. An increase in value of each dependent variable will either positively or negatively impact the wage of a person. Even though ‘height’ variable is statistically significant and negatively correlated with the constant value, this variable is perhaps not economically significant.

4.  What do the coefficients of indicator (dummy) variables like the MBA variable on Page 7 mean?

Coefficients of indicator (dummy) variables are estimated effects of unit changes in dependent variables on a dependent variable (wage in this case).

5.  What are the advantages of mean centering continuous variables? See page 8 of handout.

Mean centering will make the constant term easier to interpret. By mean centering all continuous independent variables, a negative constant value becomes positive and sometimes it is easier to interpret the constant value with a positive term.

Additionally, some independent variables do not make sense to have zero values. By mean centering all continuous independent variables, you are removing ‘zero value’ issues for some independent variables without changing coefficients of independent variables.

6.  How do you interpret the coefficient of “interaction terms?” See example on Pages 9 - 10.

The effect of one independent variable on a dependent variable is different for different values for another independent variable. For example, the coefficient of ‘c.male#c.education’ interaction variable means that the effect of being male on the wage is different for different values for education level. When a person is male and has education level 1, the effect on the wage is additional $974. If the education level is 2, the effect is $1,948 instead.

7.  How do you analyze ordinal independent variables? See examples on Pages 11 and 12.

        Ordinal variables provide either a 0 or 1 value to either be included or excluded from the regression. If included, its value is 1 and the coefficient is added/subtracted in the regression equation. For example, if a person in the study on page 11 had a FT MBA, 28525.27 would be added to the equation (and thus the value of the dependent variable). The base of having no MBA is included in the constant. Ordinal variables allow for categorical or yes/no variables to be included in a regression.

8.  What are the advantages of panel data methods we discussed in class? Understand how to interpret the results shown on Page 12.

        More accurate inference of model parameters. Greater capacity to capture the complexity of human behavior (i.e. no two people are the same, and thus an MBA will affect people differently).

        The results on page 12:

                The constant is the base wage value excluding all regression variables. Education, experience, and iqtest positively affect wages. The male variable is ordinal, and positively affects wage. Height negatively affects wages, but is economically insignificant. Having any MBA positively affects wages, however having a FT MBA vs a PT MBA is not economically or statistically significant.

Read: Class Handout (Part II) CHETSI AND NIRAV

1.          How do you interpret the interactions between two continuous variables? See page 2.

        Example from page: interaction term of experience*education

        The interaction term tests the strength of the relationship between experience and wages based on education. The term is testing whether the effect of experience on wages is different at various levels of education (i.e. Is the effect of experience on wages greater for someone who has bachelor’s vs no degree). This term slightly changes the interpretation of the coefficients for experience. To see the total effect on wages from experience, you have to look at the coefficient of experience alone (Beta1) plus the coefficient of the interaction term (Beta2*education). The second coefficient can also be seen as the additional change in the slope of the line based on education level.

2.          How do you estimate the effect of different groups on the intercept term? See page 3.

        The different groups that are being referred to are the different regions that are converted to dummy variables. When running dummy variables for a category of groups, one will be the “base case.” Each coefficient for the dummy variable shows the effect of that particular region above and beyond the “base case.” For example, in that example, the base region is 6-Mountain. So the alpha coefficient for 1-South shows the impact of the South region additional to Mountain region. If the coefficient for South is positive, you can interpret that as wages for someone from the South is expected to be more than someone from the Mountain region, provided all other variables are held constant. (Someone fact check me on this, but I think the p-value is for whether the difference between South and Mountain is statistically significant ---not sure).