Econ 107 Lab Exercise 3: Non-Linear Regression Analysis

Ran Wang

One Important Thing...

  • 1 The due data is this Thursday, May 25th by 11:59PM.

Reviews of Nonlinear Models

Linear Regression

\[ y_t=c+\alpha x_t \]

change

\[ y_1=c+\alpha x_1, y_0=c+\alpha x_0 \]

\[ y_1-y_0=\alpha (x_1-x_0) \]

\[ \Delta y~(increase~in~y)=\alpha \Delta x~(increase~in~x) \]

explaination

\( x \) changes by one unit in value, \( y \) will change by \( \alpha \) units in value.

Reviews of Nonlinear Models

Log-Linear Regression

\[ log(y_t)=c+\alpha x_t \]

change

\[ log(y_1)=c+\alpha x_1, log(y_0)=c+\alpha x_0 \]

\[ log(y_1)-log(y_0)=\alpha (x_1-x_0) \]

\[ \Delta y/y=\alpha \Delta x \]

explaination

\( x \) changes by one unit in value, \( y \) will change by \( \alpha \) units in percentage.

Reviews of Nonlinear Models

Linear-Log Regression

\[ y_t=c+\alpha log(x_t) \]

change

\[ y_1=c+\alpha log(x_1), y_0=c+\alpha log(x_0) \]

\[ y_1-y_0=\alpha (log(x_1)-log(x_0)) \]

\[ \Delta y=\alpha \Delta x/x \]

explaination

\( x \) changes by one unit in percentage, \( y \) will change by \( \alpha \) units in value.

Reviews of Nonlinear Models

Log-log Regression

\[ log(y_t)=c+\alpha log(x_t) \]

change

\[ log(y_1)=c+\alpha log(x_1), log(y_0)=c+\alpha log(x_0) \]

\[ log(y_1)-log(y_0)=\alpha (log(x_1)-log(x_0)) \]

\[ \Delta y/y=\alpha \Delta x/x \]

explaination

\( x \) changes by one unit in percentage, \( y \) will change by \( \alpha \) units in percentage.

Reviews of Nonlinear Models

Polynomial Regression

\[ y_t=c+\alpha x_t+\beta x_t^2 \]

change

\[ y_1=c+\alpha x_1+\beta x_1^2, y_0=c+\alpha x_0+\beta x_0^2 \]

\[ y_1-y_0=\alpha (x_1-x_0)+\beta(x_1^2-x_0^2) \]

\[ \Delta y=\alpha (x_1-x_0)+\beta(x_1+x_0)(x_1-x_0) \]

\[ \Delta y=(\alpha +\beta(x_1+x_0))(x_1-x_0) \]

\[ \Delta y=(\alpha +\beta(x_1+x_0))\Delta x \]

Reviews of Nonlinear Models

Polynomial Regression

\[ \Delta y=(\alpha +\beta(x_1+x_0))\Delta x \]

explaination

\( x \) changes by one unit in value, \( y \) will change by \( (\alpha +\beta(x_1+x_0)) \) units in value (The coefficient is a function of \( x_1,x_0 \)).

Reviews of Nonlinear Models

Polynomial Regression with Product Term

\[ y_t=c+\alpha x_t+\beta z_t+\gamma z_t\times x_t \]

change

Fix \( z \)

\[ y_1=c+\alpha x_1+\beta z_0+\gamma z_0\times x_1 \]

\[ y_0=c+\alpha x_0+\beta z_0+\gamma z_0\times x_0 \]

\[ y_1-y_0=\alpha(x_1-x_0)+\gamma z_0(x_1-x_0) \]

\[ \Delta y=(\alpha+\gamma z_0)\Delta x \]

Reviews of Nonlinear Models

Polynomial Regression with Product Term

\[ \Delta y=(\alpha+\gamma z_0)\Delta x \]

explaination

\( x \) changes by one unit in value, \( y \) will change by \( (\alpha +\gamma z_0) \) units in value (The coefficient is a function of \( z_0 \)).

Scatterplot in Stata

Stata Code:

twoway (scatter y x)

Stata Code (with fitted line):

twoway (scatter y x) (lfit y x)

Question Part

Does economic growth improve with more education and political stability?

Data: Growth.dta

Data Analysis:

Your dataset contains information on average growth rates for 65 countries between 1960 and 1995.

First, construct a scatterplot of \( Growth \) on \( YearsSchool \).

Question Part

Then use the outreg2 command to create a table containing:

  • (1) \( Growth \) on \( TradeShare \) and \( YearsSchool \)

  • (2) \( Growth \) on \( TradeShare \) and \( log(YearsSchool) \)

  • (3) \( Growth \) on \( TradeShare \), \( log(YearsSchool) \), \( RevCoups \), \( Assasinations \) and \( log(RGDP60) \)

  • (4) \( Growth \) on \( TradeShare \), \( log(YearsSchool) \), \( RevCoups \), \( Assasinations \), \( log(RGDP60) \) and \( TradeShare\times log(YearsSchool) \)

  • (5) \( Growth \) on \( TradeShare \), \( TradeShare^2 \), \( TradeShare^3 \), \( log(YearsSchool) \), \( RevCoups \), \( Assasinations \) and \( log(RGDP60) \)

Question Part

Write-Up

  • (a) Interpret the regression coefficients in regression (3) above.

  • (b) Using your plot, is the relationship between \( Growth \) and \( YearsSchool \) nonlinear?

  • ( c) In 1960, a country contemplated an education policy that would increase average years of schooling from 4 to 6 years. What is the predicted increase in \( Growth \) according to regressions (1) and (2)? Explain why your predictions are different.

Question Part

Write-Up

  • (d) Does the effect of \( TradeShare \) on \( Growth \) depends on the level of education (\( YearsSchool \))?

  • (e) Using regression (5), is the relationship between \( TradeShare \) and \( Growth nonlinear \)?

  • (f) In 1960, a country contemplated a trade policy that would increase the average value of \( TradeShare \) from 0.5 to 1. Use regressions (3) and (5) to predict the increase in Growth.