D.  Demand Estimation

1.  Identification problem

.

.

.

.

.

.

.

.

2.  Marketing research approaches

a.  Consumer surveys

• Ask potential customers about their buying plans

• Respondents may have difficulty answering question accurately

.

b.  Observational research

• Watch consumers buy and use products

• Use product scanners in stores and people meters in homes

• Concerns with privacy

c.  Consumer clinics

• Laboratory experiments with simulated stores

• People know they are in an experiment and may behave differently, limited sample size due to costs

.

d.  Market experiments

• Vary price and other variables in different markets

• Limited scale and duration due to costs, may be affected by outside factors

.

e.  Virtual shopping

• Have consumers shop in virtual store

.

f.  Virtual management

• Use computer simulation

• Based on theory of complexity

.

2.  Regression analysis

• Estimate an equation for demand

.

.

.

.

.

.

.

.

a.  Model specification

.

.

.

b.  Data considerations

(1)  Per capita data

.

.

.

.

.

(2)  Real data

.

.

.

.

.

c.  Specifying form

(1)  Linear

.

.

.

(2)  Nonlinear

.

.

.

d.  Estimation

Excel => Data | Data Analysis | Regression

.

.

.

.

.

.

e.  Regression statistics

(1) t-statistic

• Tests significance of parameter estimates

.

.

.

.

.

.

• Confidence interval

.

.

.

(2)  Coefficient of determination (R²)

• Measures explanatory power of the entire model

^.

^.

^.

(3)  Adjusted R²

• Takes into account number of independent variables

^.

^.

^.

(4)  F-statistic

• Tests explanatory power of the entire model

.

.

.

.

.

.

f.  Point and interval estimates

.

.

.

.

.

.

d.  Potential problems

(1)  Multicollinearity

• Some or all of the independent variables are correlated

• Leads to inefficiency (standard errors too high) and less likelihood of rejecting the hypothesis that the independent variables are insignificant

• Solve by gathering more data, using a priori information about the relationship between the variables, transforming relationship, or dropping one of the variables

.

(2)  Heteroscedasticity

• Variance of the error terms is not constant

 

.

.

.

.

.

.

.

.

.

.

• More likely with cross section data

• Results in biased standard errors => incorrect statistical tests for parameter estimates

• Transform data by using the log or dividing by the heteroscedastic variable

.

(3)  Serial or auto correlation

• Error terms are correlated

.

.

.

.

.

.

.

.

.

.

• Test using the Durbin-Watson statistic

• Use two-stage least squares to deal with the problem