Teaching Undergraduate Econometrics : A suggestion for fundamental change by Peter Kennedy

• After Stat course, most students dont understand the basic logic of classical statistics
• very little understanding of Sampling distribution concept
• Learn mechanical things: calc mean, sd, running regression and testing by remembering certain formulas
• Cram these techniques to pass and get good grades in the course
• View Statistics as branch of Mathematics and miss statiscal lens necessary to understand real world
• Miss stat lens which to view this world, allowing world to make sense
• Hendry, Peter Kennedy among others emphasizing teaching econometric not through math lens but from real world lens (statistical lens)

• Not understanding sampling distribution in econometric is a big issue
• This concept provides a unifying logic that permits
  • estimation
  • hypothesis testing
  • econometric algebraic manipulation

Context

Do Students Understand the Concept

• Garfield and Andrew Ahlgren (1988) research: not much understanding
• Peter Kennedy : Very few understand what is sampling distribution
• Students A grades in Statistics by good mechanical ability but w/o much understanding
• impressive ability to derive statistical formulas but little understanding

3 Reasons for Lack of understanding of this concept

• Its difficult: students visualize a distribution of sample data but sampling dist is high level of abstraction
• Only sampling distribtribution of mean is focused not much about regression (central focus of econometrics)
• Hubbard (1997 p.1)" We get what we assess, and if we dont assess it, we dont get it". Students are utility maximizers and they typical get higher grades without understanding it

If instructors are not aware of this deficiency they will not try to remedy it.

A suggestion for Change

• Two changes are needed
  • instructors should produce for students a good exposition of what is a sampling distribution
  • “explain how to do a Monte Carlo study” problem
  Instructors may wish to ease students into this kind of question by providing the Monte Carlo instructions and asking students to anticipate the results, as in the following example:

  1. Draw 50 x values from a uniform distribution between 3 and 12.

  2. Draw 50 z values from a standard normal distribution.

  3. Compute 50 w values as 4 - 3x + 9z.

  4. Draw 50 e values from a standard normal distribution.

  5. Compute 50 y values as 2 + 3x + 4w + 5e.

  6. Regress y on x and save the x slope coefficient estimate blI

  7. Regress y on x and w and save the x slope coefficient estimate bbl.

  8. Repeat this procedure from (d) to get 1,000 b and bb values.

  9. Compute the averages of these sets of 1,000 values to get B and BB, respectively.

  10. Compute the variances of these sets of 1,000 values to get VB and VBB, respectively. Should B or BB be closer to 3? Should VB or VBB be closer to 0?

Way Forward

• Ask students to carry out MC study
• Carefully evaluate MC description of students as small mistake & vagueness reflect bad understanding
• Begin with easy questions finding the variance of sample mean statistics
• Biggest challenge is how to simulate DGP
• In class development of a step-by-step set of MC instructions can be effective means of generating class participation
• Instructors need to spend considerable time on sampling distributions and MC studies
• Advanced estimation or testing techniques do not mean much to students if they do not understand the fundamental principles lie behind them.

Many instructors including Peter Kenneday have given up most mathematical derivations. Derivation of OLS and prove that it is BLUE without understanding the concept of a sampling distribution is not thoroughly understood.