Correlation There is a (+/-) and significant correlation between x1 and x2

Simple Linear Regression β0 We would expect y= β0 when x1=0

β1 As x1 increases by one unit, it is associated with a β1 increase/decrease in y
Standardized A one standard deviation increase in x1 is associated with one β1 standard deviation
coefficients increase/decrease in y
R2 x% of the variance in y was explained by a linear relationship with x1
Multiple Linear Regressions β0 We would expect y= β0 when all x variables are 0
β1 As x1 increases by one unit, it is associated with a β1 increase/decrease in y holding
constant all other x variables
Standardized A one standard deviation increase in x1 is associated with one β1 standard deviation
coefficients increase/decrease in y holding constant all other x variables
Simple Linear Regression β0 The predicted value of y when x1=0 is equal to β0
with Categorical Variables
β1 As x1 increases by one categorical unit, it will result in a β1 increase/decrease in y
Multiple Linear Regression β0 The predicted value of y when all x variables are 0 is equal to β0
with Categorical Variables
β1 As x1 increases by one categorical unit, it will result in a β1 increase/decrease in y
holding constant all other x variables
β2 As xk increases by one categorical unit, it will result in a βk increase/decrease in y
holding constant all other x variables
Interaction As x1 changes, it is associated with a greater increase/decrease in y if x2=1 (assuming x2
is a binary or yes/no variable, like single parent household status)
Multiple Linear Regression Definitions of β 0 is the population intercept, which represents the expected reading/math achievement
with Multiple Categorical Terms
for a public urban school
Variables (assume x1 is school
β 1 is the population coefficient on private, which is the difference in expected reading
type and x2 is
urbanicity) /math achievement between a private and a public school holding fixed urbanicity
(controlling for whether or not the school is urban, suburban, or rural)
β 2 is the population coefficient on suburban, which is the difference in expected
reading/math achievement between a suburban and an urban school, holding constant
whether the school is public or private
β 3 is the population coefficient on rural, which in this case, is the difference in
expected reading/math achievement between a rural and an urban school, holding
constant whether the school is public or private
 i represents the random population error term
Hypothesis Testing H 0 : β 2 =β3 =0 after controlling for whether the school is private or public
 Ha: at least one of the regression coefficients (β2, β3) do not equal 0 after controlling for
whether the school is public or private
Definition of Terms In addition to the above…
(Interactions)  4 represents the interaction effect between the suburban and private categories
 5 represents the interaction effect between the rural and private categories
Hypothesis Testing H 0 : β 4 =β 5 =0
H a : At least one of the regression coefficients , β 4 and β 5 does not equal 0
Interpretation Pay attention to F-change test
Two-Way ANOVA Definition of Terms Y ijk =μ+α i( x1 )+ β j (x 2)+( αβ )ij + ε ijk
μ=overall mean of Y across all factors and levels
α=the effect of x1 in the population
β=the effect of x2 in the population
α β=the interaction of x1 and x2
εijk=represents the random population error term
i indexes the level of x1
j indexes the level of x2
k indexes the kth observation in cell ij
Hypothesis Testing H0=μurban=μsuburban=μrural or H0=αurban=αsuburban=αrural=0 after taking into account β j (x 2 )
for Main Effects Ha=at least two of the means μurban=μsuburban=μrural are different
Inference Examples From these data urbanicity does not appear to have a statistically significant relationship
for Main Effects with the y variable, holding fixed whether the school is public/private because the p
value exceeds .05. (No main effect of urbanicity).
However, private schools have a statistically significant relationship with y, holding
constant urbanicity because the p value is less than .05. (Significant main effect of
school type).
Hypothesis Testing H0: α β=0
for Interactions H0: α β0

Inference for It either is statistically significant or it isn’t. If it is not significant, drop the interaction
Interaction Effects term and re-run the ANOVA, only testing the main effects.
Logit Functions Interpretation Expβ=Odds
( exp β−1 )
A one unit change in x raises the odds of y by %
100
Holding all other variables constant, a women is 34% more likely to vote Democrat
Dollars are such a fine grain measure, that it is difficult to see the effect of a one dollar
change. Therefore, increasing the measure to thousands of dollars makes it easier to see
the effect of income.