From BluWiki

[edit] Class notes

[edit] 3/28/2006

• 2-way ANOVA
  • quantitative outcome measure & two nominal predictor variables
• Regression
  • quantitative outcome measure and at least one predictor is interval or ratio (quantitative)
  • Must use compute statement to create interaction term

compute sizesq=file_siz**2.
exe.
^creates sizesq as the square of file_siz.

• Chi-Square
  • All variables are nominal

[edit] Multiple Regression

• Must square and cube predictor variable using the "compute" function
• Once you have X, X^2, and X^3, then simply use the regression function (pressing next each time)
• Analyze > Regression > Linear

When reporting answer:

• Give initial r-squared
• Then the RSquare change and its statistics
• Then give model R squared and its statistics

[edit] 3/30/2006

• if the question says forecast something, then you give two equations w/ upper & lower confidence intervals
• If it just say interpret something, then just give the model
  • Note: if a higher order term is significant, all the components of the term must be retained in the model

[edit] What do you need to talk about for hw?

• Regression
  • Model r squared & degrees of freedom (regression, residual)
  • R-square of chagne, sig. of change & degrees of freedom for each new variable
• You only need an interaction if you have two independent terms

[edit] 4/4/2006

• Logistic regression
• Analyze > regression > Binary logic
  • dependent var
  • covariate
  • define categorical variable... click first, change
    • save ... prob / group membership
    • options > classification nots, corr of covariance int

[edit] 4/13/2006

Data transformations

• Log transformation (most common)
  • You can log them, or use base 10 (not sure) - just make sure you're doing the same thing in the entire transformation?
• Determine what you need to transform by looking at scatter plot of the data (note you can log x & y)

Compute

LOG_INC = LN(INCOME).

SQR_INC = SQRT(INCOME).

INV_INC = INV(INCOME).

Lag:

Compute
Lag_var = lag(var).
Exe.

Graphing:

Graph > Interactive > Line.  enter X & Y

Dummy var:

compute
december =0.
if (month = 12) december =1.
exe.

[edit] =4/20/2006

[edit] Components of Time Series

• Try to use "real" external variables when you can
• Trend - looking at it to see what the trend is...
• Inertia - use lag variable
• Seasonality - test seasonality by dummy coding seasons
• Cyclicality - haven't really dealt w/ this

[edit] =TEST

• Two ANOVAs, 1 time series, and two regressions < last regression you're going have to read the question and "think" what is going on? You'll have to create dummy vars.