The final exam will be comprehensive. You should review the material
we covered for exam 1 and exam
2 . You will need a calculator for this exam
I. Regression (Con't)
A. Variations on OLS
1. dummy variables
2. transforming non-linear relations to use OLS
a. quadratic terms
b. natural logs
B. R-squared
1. coefficient of determination
2. low R-squared-- lots of unexplained variation
in Y
II. Regression Problems
A. regression does not show causality
B. Data collection must be methodologically sound
1. random sample, representative of population
2. avoid bias in survey questions
3. results are sensitive to extreme values (outliers)
C. Statistical Issues
1. model must be correctly specified
a. missing variables
b. spurious relations
2. multicollinearity
3. heteroskedasticity
4. autocorrelation
III. Communicating Statistics
A. Graphing
1. use of scale, units
2. sample, time period
3. graphs can be a surprisingly powerful analytic
tool
B. Problems with Statistical Reporting
1. publication bias
a. "positive" results
b. sponsor
2. conclusions may not reflect statistical results
3. results may not be statistically significant
- is a confidence level
or margin of error reported?
4. results may not have real world significance
5. biased samples or biased surveys will give meaningless
results
- is methodology reported?
C. Experimental Design
1. study and control groups
a. random
b. blind
2. measurement needs to be double-blind
3. results from poorly designed experiments are
not meaningful
a. placebo effect
b. unconscious bias
IV. Forecasting
A. Use estimated coefficients to forecast Y, given new X variables
B. Conditional forecasts
C. Systems of Equations
D. Forecasting issues & problems
a. linear or non-linear equation
b. predicting outside range of sample
c. predicting turning points is difficult
d. forecast does not account for unexpected shocks
V. Time series
A. Trends
1. growth over time
2. contraction
B. Fluctuation
1. seasonal
2. cyclical
3. random
4. smoothing techniques
a. moving average
b. exponential
smoothing
c. seasonal
adjustment
C. Time Series Regression Analysis
1. estimating trend lines
Y = a + bT
2. univariate procedures
a. autoregressive process
-use lagged values of Y as explanatory variables
b. ARIMA
-autoregressive integrated moving average
3. multivariate procedures
a. variables that trend
together
b. compare change in Y to
change in X
c. cointegration