Lecture 10: Alternatives to OLS with

limited dependent variables, part 1

 PEA vs APE
 Logit/Probit
 PEA vs APE

 PEA: partial effect at the average

 The effect of some x on y for a hypothetical case with
sample averages for all x’s.
 This is obtained by setting all Xs at their sample
mean and obtaining the slope of Y with respect to
one of the Xs.
 APE: average partial effect
 The effect of x on y averaged across all cases in the
sample
 This is obtained by calculating the partial effect for all
cases, and taking the average.
 PEA vs APE: different?

 In OLS where the independent variable is

entered in a linear fashion (no squared or
interaction terms), these are equivalent. In fact, it
is an assumption of OLS that the partial effect of
X does not vary across x’s.
 PEA and APE differ when we have squared or
interaction terms in OLS, or when we use
logistic, probit, poisson, negative binomial, tobit
or censored regression models.
 PEA vs APE in Stata

 The “margins” function can report the PEA or the

APE. The PEA may not be very interesting
because, for example, with dichotomous
variables, the average, ranging between 0 and
1, doesn’t correspond to any individuals in our
sample.
 . “margins, dydx(x) atmeans” will give you the
PEA for any variable x used in the most
recent regression model.
 . “margins, dydx(x)” gives you the APE
PEA vs APE

 In regressions with squared or interaction

terms, the margins command will give the
correct answer only if factor variables have
been used
 http://www.public.asu.edu/~gasweete/crj604/
misc/factor_variables.pdf
 APE and PEA Stata exercise
 Use the midterm nlsy data:
 http://www.public.asu.edu/~gasweete/crj604/midterm/mid11nlsy.dta
 Calculate a model of relationship quality using just
male, antisocial peers, and their interaction term
 Obtain:
 PEA for male and antisocial peers (create
interaction term using factor variables)
 APE for male and antisocial peers (use a
created interaction term)
 Limited dependent variables
 Many problems in criminology require that we analyze
outcomes with very limited distributions
 Binary: gang member, arrestee, convict, prisoner
 Lots of zeros: delinquency, crime, arrests
 Binary & continuous: criminal sentences (prison or
not & sentence length)
 Censored: time to re-arrest
 We have seen that large-sample OLS can handle
dependent variables with non-normal distributions.
However, sometimes the predictions are nonsensical,
and often they are hetoroskedastic.
 Many alternatives to OLS have been developed to deal
with limited dependent variables.
 Review of problems of the LPM
 Recall, the Linear probability Model uses OLS with a
binary dependent variable. Each coefficient represents
the expected change in the probability that Y=1, given a
one point change in each x.
 While it is easy to interpret the results, there are a few
problems.
 Nonsensical predictions: above 1, below 0
 Heteroskedasticity
 Non-normality of errors: for any set of x’s the error
term can take on only two values: y minus yhat, or
negative yhat
 Linearity assumption: requiring that X has equal
effect across other Xs is not practical. There are
diminishing returns approaching 0 or 1.
 Binary response models (logit, probit)
 There exists an underlying response variable Y* that
generates the observed Y (0,1).
 Binary response models (logit, probit)
 Y* is continuous by unobserved. What we observe is a
dummy variable Y, such that:

 When we incorporate explanatory variables into the

model, we think of these as affecting Y*, which in turn,
affects the probability that Y=1.
 Binary response models (logit,
probit)

 This leads to the following relationship:

E (Y )  P(Y  1)  P(Y *  0)
 We generally choose from two options for modeling Y*
 normal distribution (probit)
 logistic distribution (logit)
 In each case, using the observed Xs, we model the area
under the probability distribution function (max=1) up to the
predicted value of Y*. This becomes P(Y=1) or the
expected value of Y given Xs.
Probit and logit cdfs
 Probit and logit models, cont.
 Clearly, the two distributions are very similar, and they’ll
yield very similar results.
 The logistic distribution has slightly fatter tails, so it’s
better to use when modeling very rare events.
 The function for the logit model is as follows:

exp( yˆ ) *
yˆ  P( y  1) 
1  exp( yˆ )
*
 Logit model reporting
 In Stata, at least two commands will estimate the logit model
 Logit Y X reports the coefficients
 Logistic Y X reports odds ratios
 What’s an odds ratio?
 Back up, what’s an odds?
 An odds is a ratio of two numbers. The first is the
chances an event will happen, the second are the
relative chances it won’t happen.
 The odds that you roll a 6 on a six-sided die is 1:5,
or .2
 The probability that you roll a 6 is 1/6 or about .167
 Logit model reporting
 Probabilities and odds are directly related. If p is the
probability that an event occurs, the odds are p/(1-p)
 P=1, odds=undefined
 P=.9, odds=.9/.1=9
 P=.5, odds=.5/.5=1
 P=.25, odds=.25/.75=1/3
 Likewise, if the odds of an event happening is equal to q,
the probability p equals q/(1+q)
 Odds=5, p=5/6=.833
 Odds=1.78, p=1.78/2.78=.640
 Okay, now what’s an odds ratio? Simply the ratio
between two odds.
Logit model reporting

 Suppose we say that doing all the homework and

reading doubles the odds of receiving an A in a
course. What does this mean?
 Well, it depends on what the original odds of
receiving an A in course.
Original New odds Original p New p ∆p
odds
5 10 .83 .91 .08
1 2 .50 .67 .17
.75 1.5 .43 .60 .17
.3333 .6666 .25 .40 .15
.01 .02 .0099 .0196 .0097
 Logit model reporting

 So what does this have to do with logit model

reporting?
 Raw coefficients, reported using the “logit” command
in Stata, can be converted to odds ratios by
exponentiating them: exp(βj)
 Let’s look at an example from Sweeten (2006), a
model predicting high school graduation. Odds ratios
are reported . . .
 Nonrandom samples / missing
data
 Endogenous sample selection: based on the dependent
variable
 This biases your estimates.
 Missing data can lead to nonrandom samples as well.
 Most regression packages perform listwise deletion of all
variables included in OLS. That means that if any one of the
variables is missing, then that observation is dropped from
the analysis.
 If variables are missing at random, this is not a problem, but it
can result in much smaller samples.
 20 variables missing 2% of observations at random results
in a sample size that is 67% of the original (.98^20)
 Marginal effects in logistic
regression
 You have several options when reporting effect size
in logistic regression.
 You can stay in the world of odds ratios, and simply
report the expected change in odds for a one unit
change in X. Bear in mind, however, that this is not a
uniform effect. Doubling the odds of an event can
lead to a 17 percentage point change in the
probability of the event occurring, down to a near-
zero effect.
 You can report the expected effect at the mean of
the Xs in the sample. (margins command)
 Marginal effects in logistic
regression, cont.

 If there is a particularly interesting set of Xs, you can

report the marginal effect of one X given the set of
values for the other Xs.
 You can also report the average effect of X in the
sample (rather than the effect at the average level of
X). They are different. See Stata log.
 Goodness of fit
 Most stat packages report pseudo-r2. There are
many different formulas for psuedo-r2. Generally,
they are more useful in comparing models than in
assessing how well the model fits the data.
 We can also report the percent of cases correctly
classified, setting the threshold at p>.5, or preferably
at the average p in the sample.
 Careful though, with extreme outcomes, it’s very
easy to get a model that predicts nearly all cases
correctly without predicting the cases we want to
predict correctly.
 Goodness of fit, cont.
 For example, if only 3% of a sample is arrested, an easy way
to get 97% accuracy in your prediction is to simply predict
that nobody gets arrested.
 Getting much better than 97% accuracy in such a case can
be very challenging.
 The “estat clas” command after a logit regression gives us
detailed statistics on how well we predicted Y.
 Specificity: true negatives/total negatives, % of negatives
identified, goes down as false positives go up
 Sensitivity: true positives/total positives, % of positives
identified, goes down as false negatives go up
 It also gives us the total correctly classified. All these number
change depending on the threshold used.
 Goodness of fit, cont.
 “estat clas” also gives us the total correctly classified. All
these number change depending on the threshold used.
 “lsens” shows the relationship between threshold, sensitivity
and specificity.
 “lroc” shows the relationship between the false positive rate
(X-axis) and the true positive rate (Y-axis). If you want to
have more true positives, you need to accept more false
positives.
 “lroc” also reports area under the curve. The maximum is 1,
which is only attainable in a perfect model (100% true
positives & 0% false positives). Generally, the closer you are
to 1, the better the model is.
 Probit model
 The probit model is quite similar to the logit model in
the setup and post-estimation diagnostics. However,
the coefficients are not exponentiated and
interpreted as odds ratios.
 Rather, coefficients in probit models are interpreted
as the change in the Z-score for the normal
distribution associated with a one unit increase in x.
 Clearly, the magnitude of a change then depends on
where you begin on the normal curve, which
depends on the values of the other Xs. Also, at
extreme values, the absolute effect of changes in X
diminish.
 Logit Stata exercise
 Use the midterm nlsy data:
 http://www.public.asu.edu/~gasweete/crj604/midterm/mid11nlsy.dta
 Calculate a model of predictors of discussing marriage
using male, age, dating duration, relationship quality and
an interaction term between male and dating duration
 Report the odds ratio for male when dating duration is 2
years
 Report the odds ratio for dating duration for females
 Report the PEA/APE for age and male
 What are the sensitivity and specificity using .5 as the
threshold. How does this change when the sample
mean for discussing marriage is used?
 Is this a good model for predicting discussing marriage?
Use the psuedo-r2 and lroc graph
 Next time:

Homework: 17.2, C17.2 change question iv to “”estimate

the average partial effect and the partial effect at the
average discrimination effect for the logit and probit
models.” and the logit stata exercise on the previous
slide.

Re-Read: Wooldridge Chapter 17, look over Bushway et

al., 2007, Smith & Brame, 2003 (blackboard)