You are on page 1of 5

ANALYTICS REPORT

TO: SALARY REPORTER

FROM: OLIVIA LAWSON

SUBJECT: NON-LINEAR REGRESSION

DATE: 29 OCTOBER 2020

Introduction

In help to further understand how starting salary can impact one’s mid-career salary, we
evaluated 147 state schools and their students’ salaries. The data below was computed by
running a linear regression on the students’ data. After running the regression, it was found that
the Log-Log model gives us the most accurate depiction for how starting salary affects mid-
career salary due to its high r squared value and low standard error value.

Data Analysis

Quad
Regression Equation- ^ MCMS=−9307.61+2.04 ( SMS )−7.53 E - .07( SMS )2
This model concaves down due to the negative value in front of the Starting Median
Salary squared.
bSMS= As Starting Median Salary increases by $1, Median Mid-Career Salary decreases
by $2.04 on average and all else constant.
Marginal Effect ¿ 2.04+ 2 (−7.53 ) SMS

Lin-Log
Regression Equation- ^
MCMS=−872114.3121+ 89002.87641 ln ⁡(SMS )
This model concaves up due to the positive value in front of the natural log of Starting
Median Salary.
bln(SMS)= As Starting Median Salary increases by 1%, Median Mid-Career Salary
increases by $89002.8764 on average and all else constant.
Log-Lin
Regression Equation- ln ( ^
MCMS )=10.2061+2.42 E -.05(SMS)
This model concaves up due to the positive value in front of Starting Median Salary.
bSMS = As Starting Median Salary increases by $1, Median Mid-Career Salary will
increase by 2.42E-.03 % on average and all else constant.
Log-Log
Regression Equation- ln ( MCMS )=−0.4599+1.0976 ln ( SMS )
This model concaves up due to the positive value in front of the natural log of Starting
Median Salary.
bln(SMS)= As Starting Median Salary increases by 1%, Median Mid-Career Salary
increases by 1.0976% on average and all else constant.

Best Model
After calculation we can confidently say that Log-Log is the best model because of its
higher R squared value when compared to the quadratic model. We had to calculate a
new R squared value for Log-Log because the two models used different y variables,
which does not allow us to make a comparison. After narrowing down the best model
between Quad v. Lin-Log and Log-Lin v Log-Log, the new R squared value was
calculated to make an equal comparison.

Log-Log

R squared = 0.75738; We are 75.74% of the way toward perfectly predicting the natural
log of Median Mid-Career Salary.
Standard Error = 0.05864; On average, our predictions of natural log of Median Mid-
Career Salary are off by an average of 0.0586 log dollars using this model.

Prediction with starting salary of $55,000-


ln ( MCMS )=−0.4599+1.0976 ln ( 55000 )
ln ( MCMS )=11.5205
Conclusion
The Log-Log model gave us the most accurate depiction of how starting salary can affect mid-
career salary. This was discovered by running a linear regression, narrowing down the best
models based off r squared and standard error values, and by lastly computing a new r squared
value to allow for accurate comparison. If there are any questions regarding the computation of
these models, please feel free to reach out to Olivia Lawson via email at:
olawson@email.arizona.edu .
Appendix

Quadratic Residual

Lin-Log Residual
Log-Lin
Log-Log

Adjusted R^2 Interpretation

You might also like