Regression-Logistic-4.pdf

The coefficients of the multiple regression model are
estimated using sample data with k independent
variables
• Interpretation of the Slopes: (referred to as a Net
Regression Coefficient)
– b1=The change in the mean of Y per unit change in X1,
taking into account the effect of X2 (or net of X2)
– b0 Y intercept. It is the same as simple regression.
ki
k
2i
2
1i
1
0
i
X
b
X
b
X
b
b
Ŷ +
+
+
+
= 
Estimated
(or predicted)
value of Y
Estimated slope coefficients
Estimated
intercept
2

Binary Logistic Regression
• Binary logistic regression is a type of
regression analysis where the dependent
variable is a dummy variable (coded 0, 1)
• Why not just use ordinary least squares?
Y = a + bx
– You would typically get the correct answers in
terms of the sign and significance of
coefficients
– However, there are three problems
^
3

OLS on a dichotomous dependent variable:
Yes = 1
No = 0
X = Income
Y = Support
Privatizing
Social
Security 1 10
4

– However, there are three problems
1. The error terms are heteroskedastic (variance of
the dependent variable is different with different
values of the independent variables
2. The error terms are not normally distributed
3. And most importantly, for purpose of
interpretation, the predicted probabilities can be
greater than 1 or less than 0, which can be a
problem for subsequent analysis.
5

• The “logit” model solves these problems:
– ln[p/(1-p)] = a + BX
or
– p/(1-p) = ea + BX
– p/(1-p) = ea (eB)X
Where:
“ln” is the natural logarithm, logexp, where e=2.71828
“p” is the probability that Y for cases equals 1, p (Y=1)
“1-p” is the probability that Y for cases equals 0,
1 – p(Y=1)
“p/(1-p)” is the odds
ln[p/1-p] is the log odds, or “logit” 6

• Logistic Distribution
• Transformed, however,
the “log odds” are linear.
ln[p/(1-p)]
P (Y=1)
x
x
7

• The logistic regression model is simply a
non-linear transformation of the linear
regression.
• The logistic distribution is an S-shaped
distribution function (cumulative density
function) which is similar to the standard
normal distribution and constrains the
estimated probabilities to lie between 0
and 1.
9

With the logistic transformation, we’re fitting
the “model” to the data better.
• Transformed, however, the “log odds” are
linear.
P(Y = 1) 1
.5
0
X = 0 10 20
Ln[p/(1-p)]
X = 0 10 20
10

• You’re likely feeling
overwhelmed, perhaps
anxious about
understanding this.
• Don’t worry, coherence is
gained when you see
similarity to OLS regression:
1. Model fit
2. Interpreting coefficients
3. Inferential statistics
4. Predicting Y for values of the
independent variables (the
most difficult, but we’ll make it
easy)
11

Review & Summary
• In logistic regression, we predict Z, not p,
because of Z’s convenient mathematical
properties
• Z is a linear function of the predictors,
and we can translate that prediction into a
probability.
12

Logistic regression predicts the
natural logarithm of the odds
• The natural log of the odds is called the
“logit” =“Z”
• Formula:
Z = log (p/1-p) = B0 + B1.X1 + B2.X2 +
B3.X3 . . . e
13

• B’s in logistic regression are analogous to
b’s in OLS
• B1 is the average change in Z per one unit
increase in X1, controlling for the other
predictors
• We calculate changes in the log odds of
the dependent variable, not changes in the
dependent variable (as in OLS).
14

Interpreting logistic regression results
• In SPSS output, look for:
1) Model chi-square (equivalent to F)
2) WALD statistics and “Sig.” for each B
3) Logistic regression coefficients (B’s)
4) Exp(B) = odds ratio
15

Interpreting logistic coefficients
• Identify which predictors are significant
by looking at “Sig.”
• Look at the sign of B1
* If B1 is positive, a unit change in x1 is
raising the odds of the event happening,
after controlling for the other predictors
* If B1 is negative, the odds of the event
decrease with a unit increase in x1.
16

Interpreting the odds ratio
• Look at the column labeled Exp(B)
Exp(B) means “e to the power B” or eB
Called the “odds ratio” (Gr. symbol: Ψ)
e is a mathematical constant used as the
“base” for natural logarithms
• In logistic regression, eB is the factor by
which the odds change when X increases by
one unit.
17

Interpreting the odds ratio
• New odds / Old odds = eB = odds ratio
• e.g. if the odds-ratio for EDUC is 1.05, that
means that for every year of education, the
odds of the outcome (e.g. voting) increase by a
factor of 1.05.
• Odds ratios > 1 indicate a positive relationship
between IV and DV (event likely to occur)
• Odds ratios < 1 indicate a negative relationship
between IV and DV (event less likely to occur)
18

Let’s come up with an
example … run it … and
interpret it …

• A researcher is interested in the likelihood of
gun ownership in the US, and what would
predict that.
• She uses the GSS to test the following
research hypotheses:
1. Men are more likely to own guns than are women
2. Older people are more likely to own guns
3. White people are more likely to own guns than are
those of other races
4. More educated people are less likely to own guns
20

• Variables are measured as such:
Dependent:
Havegun: no gun = 0, own gun(s) = 1
Independent:
1. Sex: men = 0, women = 1
2. Age: entered as number of years
3. White: all other races = 0, white =1
4. Education: entered as number of years
SPSS: Analyze  Regression  Binary Logistic
Enter your variables and for output below, under
options, I checked “iteration history”
21

SPSS Output: Some descriptive information first…
22

Goodness-of-fit statistics for new model come next…
Test of new model vs. intercept-
only model (the null model), based
on difference of -2LL of each. The
difference has a X2 distribution. Is
new -2LL significantly smaller?
The -2LL number is “ungrounded,” but it has a χ2
distribution. Smaller is better. In a perfect model, -2 log
likelihood would equal 0.
These are attempts to
replicate R2 using information
based on -2 log likelihood,
(C&S cannot equal 1)
-2(∑(Yi * ln[P(Yi)] + (1-Yi) ln[1-P(Yi)])
Assessment of new model’s
predictions
23

Goodness-of-fit statistics for new model come next…
17.6% of the variance in gun
ownership is explained by
gender, age, race, and
education
24

Remember When Assessing
Predictors, The Odds Ratio or Exp(b)…
• Indicates the change in odds resulting
from a unit change in the predictor.
– OR > 1: Predictor ↑, Probability of outcome
occurring ↑.
– OR < 1: Predictor ↑, Probability of outcome
occurring ↓.
predictor
the
in
change
unit
a
before
Odds
predictor
the
in
change
unit
a
after
Odds
b
Exp =
)
(
25

Interpreting Coefficients…
ln[p/(1-p)] = a + b1X1 + b2X2 + b3X3 + b4X4
b1
b2
b3
b4
a
Being male, getting older, and being white have a positive effect on likelihood of
owning a gun. On the other hand, education does not affect owning a gun.
We’ll discuss the Wald test in a moment…
X1
X2
X3
X4
1
eb
Which b’s are significant?
26

Each coefficient increases the odds by a multiplicative amount, the amount is eb. “Every
unit increase in X increases the odds by eb.”
In the example above, eb = Exp(B) in the last column.
New odds / Old odds = eb = odds ratio
For Female: e-.780 = .458 …females are less likely to own a gun by a factor of .458.
Age: e.020=1.020 … for every year of age, the odds of owning a gun increases by a
factor of 1.020.
White: e1.618 = 5.044 …Whites are more likely to own a gun by a factor of 5.044.
Educ: e-.023 = .977 …Not significant
27

Each coefficient increases the odds by a multiplicative amount, the amount is eb. “Every
unit increase in X increases the odds by eb.”
For Sex: e-.780 = .458 … If you subtract 1 from this value, you get the proportion increase
(or decrease) in the odds caused by being male, -.542. In percent terms, odds of
owning a gun decrease 54.2% for women.
Age: e.020=1.020 … A year increase in age increases the odds of owning a gun by 2%.
White: e1.618 = 5.044 …Being white increases the odd of owning a gun by 404%
Educ: e-.023 = .977 …Not significant
28

Here’s another example …
and another way to interpret
the results

Equation for Step 1
We can say that the odds of a patient who is treated being cured are 3.41
times higher than those of a patient who is not treated, with a 95% CI of
1.561 to 7.480.
The important thing about this confidence interval is that it doesn’t cross 1
(both values are greater than 1). This is important because values greater
than 1 mean that as the predictor variable(s) increase, so do the odds of (in
this case) being cured. Values less than 1 mean the opposite: as the
predictor increases, the odds of being cured decreases.
See p 288 for an
Example of using
equation to compute
Odds ratio.
30

Output: Step 1
Removing Intervention from the model would have a significant effect on
the predictive ability of the model, in other words, it would be very bad to
remove it.
31

The test you choose depends on level of measurement:
Independent Variable Dependent Variable Test
Dichotomous Interval-Ratio Independent Samples t-test
Dichotomous
Nominal Nominal Cross Tabs
Dichotomous Dichotomous
Nominal Interval-Ratio ANOVA
Dichotomous Dichotomous
Interval-Ratio Interval-Ratio Bivariate Regression/Correlation
Dichotomous
Two or More…
Interval-Ratio
Dichotomous Interval-Ratio Multiple Regression
Interval-Ratio
Dichotomous Dichotomous Binary Logistic Regression
33

The Multiple Regression Model building
Idea: Examine the linear relationship between
1 dependent (Y) & 2 or more independent variables (Xi)
ε
X
β
X
β
X
β
β
Y ki
k
2i
2
1i
1
0
i
+
+
+
+
+
= 
Multiple Regression Model with k Independent Variables:
Y-intercept Population slopes Random Error
34

• So what are natural logs and exponents?
– If you didn’t learn about them before this class, you
obviously don’t need to know it to get your degree …
so don’t worry about it.
– But, for those who did learn it, ln(x)=y is the same as:
x=ey
READ THE ABOVE
LIKE THIS: when you see “ln(x)” say “the
value after the equal sign is the
power to which I need to take e
to get x”
so…
y is the power to which you
would take e to get x 35

• So… ln[p/(1-p)] = y is same as: p/(1-p) = ey
READ THE ABOVE
LIKE THIS: when you see “ln[p/(1-P)]” say
“the value after the equal sign is
the power to which I need to take
e to get p/(1-p)”
so…
y is the power to which you
would take e to get p/(1-p)
36

• So… ln[p/(1-p)] = a + bX is same as: p/(1-p) = ea + bX
READ THE ABOVE
LIKE THIS: when you see “ln[p/(1-P)]” say
“the value after the equal sign is
the power to which I need to take
e to get p/(1-p)”
so…
a + bX is the power to which you
would take e to get p/(1-p)
37

• Recall that OLS Regression used an “ordinary least squares”
formula to create the “linear model” we used.
• The Logistic Regression model will be constructed by an iterative
maximum likelihood procedure.
• This is a computer dependent program that:
1. starts with arbitrary values of the regression coefficients and constructs
an initial model for predicting the observed data.
2. then evaluates errors in such prediction and changes the regression
coefficients so as make the likelihood of the observed data greater
under the new model.
3. repeats until the model converges, meaning the differences between
the newest model and the previous model are trivial.
• The idea is that you “find and report as statistics” the parameters
that are most likely to have produced your data.
• Model and inferential statistics will be different from OLS because
of using this technique and because of the nature of the dependent
variable. (Remember how we used chi-squared with
classification?)
38

• So in logistic regression, we will take the “twisted”
concept of a transformed dependent variable equaling
a line and manipulate the equation to “untwist” the
interpretation.
• We will focus on:
1. Model fit
2. Interpreting coefficients
3. Inferential statistics
4. Predicting Y for values of the independent variables (the most
difficult)—the prediction of probability, appropriately, will be an
S-shape
• Let’s start with a research example and SPSS
output…
39

SPSS Output: Some descriptive information first…
Maximum likelihood process stops at
third iteration and yields an intercept
(-.625) for a model with no
predictors.
A measure of fit, -2 Log likelihood is
generated. The equation producing
this:
-2(∑(Yi * ln[P(Yi)] + (1-Yi) ln[1-P(Yi)])
This is simply the relationship
between observed values for each
case in your data and the model’s
prediction for each case. The
“negative 2” makes this number
distribute as a X2 distribution.
In a perfect model, -2 log likelihood
would equal 0. Therefore, lower
numbers imply better model fit.
40

Originally, the “best guess” for each
person in the data set is 0, have no gun!
This is the model for log
odds when any other
potential variable equals
zero (null model). It
predicts : P = .651, like
above. 1/1+ea or 1/1+.535
Real P = .349
If you added
each…
41

Next are iterations for our full model…
42

• ln[p/(1-p)] = a + b1X1 + …+bkXk, the power to which you
need to take e to get:
P P
1 – P So… 1 – P = ea + b1X1+…+bkXk
• Ergo, plug in values of x to get the odds ( = p/1-p).
The coefficients can be manipulated as follows:
Odds = p/(1-p) = ea+b1X1+b2X2+b3X3+b4X4 = ea(eb1)X1(eb2)X2(eb3)X3(eb4)X4
Odds = p/(1-p) = ea+.898X1+.008X2+1.249X3-.056X4 = e-1.864(e.898)X1(e.008)X2(e1.249)X3(e-.056)X4
43

The coefficients can be manipulated as follows:
Odds = p/(1-p) = ea+b1X1+b2X2+b3X3+b4X4 = ea(eb1)X1(eb2)X2(eb3)X3(eb4)X4
Odds = p/(1-p) = e-2.246-.780X1+.020X2+1.618X3-.023X4 = e-2.246(e-.780)X1(e.020)X2(e1.618)X3(e-.023)X4
Each coefficient increases the odds by a multiplicative
amount, the amount is eb. “Every unit increase in X
increases the odds by eb.”
44

Age: e.020 = 1.020 … A year increase in age increases the odds of owning a gun by 2%.
How would 10 years’ increase in age affect the odds? Recall (eb)X is the equation component
for a variable. For 10 years, (1.020)10 = 1.219. The odds jump by 22% for ten years’ increase
in age.
Note: You’d have to know the current prediction level for the dependent variable to know if this
percent change is actually making a big difference or not!
45

Note: You’d have to know the current prediction level for the dependent variable to
know if this percent change is actually making a big difference or not!
Recall that the logistic regression tells us two things at once.
• Transformed, the “log odds” are linear.
ln[p/(1-p)]
P (Y=1)
x
x
46

We can also get p(y=1) for particular folks.
Odds = p/(1-p); p = P(Y=1)
With algebra…
Odds(1-p) = p … Odds-p(odds) = p …
Odds = p+p(odds) … Odds = p(1+odds)
… Odds/1+odds = p or
p = Odds/(1+odds)
Ln(odds) = a + bx and odds = e a + bx
so…
P = ea+bX/(1+ ea+bX)
We can therefore plug in numbers for X to get P
If a + BX = 0, then p = .5 As a + BX gets really big, p approaches 1
As a + BX gets really small, p approaches 0 (our model is an S curve)
47

For our problem, P = e-2.246-.780X1+.020X2+1.618X3-.023X4
1 + e-2.246-.780X1+.020X2+1.618X3-.023X4
For, a man, 30, Latino, and 12 years of education, the P equals?
Let’s solve for e-2.246-.780X1+.020X2+1.618X3-.023X4 = e-2.246-.780(0)+.020(30)+1.618(0)-.023(12)
e-2.246 – 0 + .6 + 0 - .276 = e -1.922 = 2.71828-1.922 = .146
Therefore,
P = .146 = .127 The probability that the 30 year-old, Latino with 12
1.146 years of education will own a gun is .127!!! Or
you could say there is a 12.7% chance.
48

Inferential statistics
are as before:
• In model fit, if χ2 test
is significant, the
expanded model
(with your variables),
improves prediction.
• This Chi-squared test
tells us that as a set,
the variables improve
classification.
49

Inferential statistics are as before:
• The significance of the coefficients is
determined by a “wald test.” Wald is χ2 with
1 df and equals a two-tailed t2 with p-value
exactly the same.
50

1. Significance test for α-level = .05
2. Critical X2
df=1= 3.84
3.To find if there is a significant slope in the population,
Ho: β = 0
Ha: β ≠ 0
4.Collect Data
5.Calculate Wald, like t (z): t = b – βo (1.96 * 1.96 = 3.84)
s.e.
6.Make decision about the null hypothesis
7.Find P-value
So how would I do hypothesis testing? An Example:
Reject the null for Male, age, and white. Fail to reject the null for education.
There is a 24.2% chance that the sample came from a population where
the education coefficient equals 0. 51

Regression-Logistic-4.pdf

More Related Content

Similar to Regression-Logistic-4.pdf

More from jiregnaetichadako

Recently uploaded

Regression-Logistic-4.pdf