Linear Probability

Response Models

Probit and Logit

Techniques for analyzing the

relationship between fixed-level
independent variables and a
dependent variable constrained to
vary between 0 and 1
 2

Key Concepts
*****
Probit & Logit

The analytic differences between linear regression and the linear probability
models Probit and Logit
Fixed v. random variable
Constrained v. unconstrained variables
The analytic differences between logistical regression and Probit & Logit
The definition of a Probit transformation
The definition of a Logit transformation
Assumptions of Probit and Logit
Maximum likelihood estimation and Probit/Logit parameters
The meaning of linearity in Probit and Logit analysis
Interpretation of the constant (a) and regression coefficient (b) in Probit analysis
Interpretation of the constant (a) and regression coefficient (b) in Logit analysis
Chi square goodness-of-fit test in Probit and Logit and its null hypothesis
Expected residual scatterplot if Probit/Logit a and b are not significant
The concept of a residual in Probit and Logit
The concept of median potency
Back-solving for the median potency from a Probit or Logit equation
95% confidence interval of the median potency
The purpose and method of logarithmic transformation in Probit and Logit
analysis
Common v. natural logarithms
Logarithms and antilogarithms
Laws of logarithms
Bivariate Probit and Logit models
Multivariate Probit and Logit models
Probit and Logit models with multiple groups and/or multiple IV(s)
The concept of a common slope in Probit/Logit multiple group analysis
Parallelism test for the homogeneity of regression slopes for factor levels

*****

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 3

Lecture Outline

An example: predicting success of juvenile

probationers

The theory of the Probit Response Model

Improving the goodness of fit of a Probit model

with logarithmic transformation

Interpreting a log transformed Probit model

The theory of the Logit Response Model

Comparison of Probit & Logit Response Models

Probit with two independent variables

Probit with multiple groups

Probit with multiple independent variables &

multiple groups

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 4

Analytic Problem

Linear regression assumes

That there are no restrictions on the values
of the IV(s), that the values are random
That the DV is continuous
And that since the IV(s) can take on any
value, the DV can range from ± ∞
The problem
What if the IV(s) has fixed levels, vis-à-vis
being random?
What if the DV is constrained to be between
0.0 & 1.0 (as in a proportion)?
Use of linear regression would lead to an
egregious violation of assumptions

The solution
Use a linear probability model such as
Probit or Logit

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 5

Database: Juvenile Probation 1

Probation Outcome Among Juvenile Shoplifters as a
Function of Months of Supervision

Succeeded Failed
Months Total N
N p N p

6 18 .47 21 .53 38

7 23 .55 19 .45 42

8 22 .61 14 .39 36

9 19 .63 11 .37 30

10 32 .66 16 .34 48

11 32 .71 13 .29 45

12 30 .75 10 .25 40

Totals 176 103 279

N = number of juveniles, p = proportion of successful juveniles

Independent variable (metric & fixed): months of supervision

Dependent variable: the proportion that succeeded
Overall success = (176) / (279) = 0.63, or 63%

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 6

Research Questions

1. Is there a significant relationship between

success on probation and the number of months
of supervision?

2. Analytic approaches:

Series of t-tests comparing differences in

proportions of successes for each of seven
supervision groups

Chi square analysis of months by

success/failure

Logistical regression: success/failure (0/1)

as a function of months

3. By how much does the proportion of successes

increase for each one month of supervision?

4. How much supervision was required to achieve a

50% success rate, 75%, 90%, etc.?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 7

Research questions ( con'd )

5. Multivariate questions:

Does the effect of the number of months of

supervision on success vary with the age of the
juvenile?

Does the effect of the number of months

supervision on success vary with the age and
gender of the juvenile?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 8

Probit: Linear Probability Model

Transform (Pi) = a + bXI

Dependency Technique
One or more metric IV(s) reduced to fixed
levels or “doses”
One binary DV, measured as the proportion
of the subjects responding to a fixed level of
the IV. DV is transformed to Z score values.
Parameters are estimated by maximum
likelihood estimation, not OLS.

Assumptions
The relationship between the levels of the
IV(s) and the DV is linear
Homogeneity of variance of the DV over
levels of the IV
Homoskedastic residuals
Absence of outliers

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 9

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 10

The Probit Model

Transform (Pi) = a + bXI

PI = Probit: the Z score associated with the
proportion of cases responding to the stimulus
(e.g. proportion successful on probation)
a = constant
b = Probit regression coefficient
Xi = Level of the independent variable (e.g.
months of supervision)

Probit transformation: Convert the proportion

responding to a level of the IV to the corresponding
Z score on a standard normal curve
Example: 22 out of 36 juveniles succeeded with 8
months supervision. (22) / (36) = 0.611

Z = + 0.28 (0.611 is approximately the

proportion in the larger area of the curve
associated with a Z score of +0.28)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 11

Insert a standard normal probability table with values

of Z from 0.0 to 0.40

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 12

Probit Transformation

22 out of 36 juveniles succeeded

with 8 months of supervision (61.1%)

Standard Normal Curve

Area = 0.611

Z = 0.00

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 13

Z = + 0.28

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 14

Probit & Logit Case Studies of

Probation Outcome Study

Model 1 Probit response model

Success = f (months of supervision)

Model 2 Log transform Probit model

Success = f (log months)

Model 3 Logit response model

Success = f (months of supervision)

Model 4 Log transform Logit model

Success = f (log months)

Model 5 Probit response model

Success = f (months & age)

Model 6 2 Probit response models

Male Success = f (months)
Female Success = f (months)

Model 7 2 Probit multivariate models

Male Success = f (months & age)
Female Success = f (months & age)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 15

Model 1
The Effect of the Number of Months of
Supervision on Probation Success

Probit = a + b (months)

Probit = -0.70946 + 0.11582 (months)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 16

Probit Model of the Effect of Months of

Supervision on Probation Success

Months of Total Number Number

Supervision of Juveniles Successful

6 38 18 (0.47)

7 42 23 (0.55)

8 36 22 (0.61)

9 30 19 (0.63)

10 48 32 (0.67)

11 45 32 (0.71)

12 40 30 (0.75)

Values in parentheses are the proportions of juveniles who were

successful.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 17

Conversion of the Proportion Successful to

Associated Z Scores

Months of Total Number Probit Z

Supervision Number of Successful Score
Juveniles (p)

6 38 18 (0.47) -0.08

7 42 23 (0.55) +0.13

8 36 22 (0.61) +0.28

9 30 19 (0.63) +0.33

10 48 32 (0.67) +0.44

11 45 32 (0.71) +0.55

12 40 30 (0.75) +0.67

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 18

Plot of the Number of Successful

Probationers as a Function of
Months of Supervision

Probation Success as a Function of

Months of Supervision
34

32

30

28

26

24

22

20

18
16
5 6 7 8 9 10 11 12 13

MONTHS

Compare this scatterplot with the Probit transform

plot on the following exhibit.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 19

Plot of Probit Z Scores as a Function of

Months of Supervision

Probit Transformed Responses

.8

.6

.4

.2

0.0

-.2
5 6 7 8 9 10 11 12 13

MONTHS

Notice that the conversion of the proportion of

successful juveniles to Probit Z scores results in a
linear relationship over time.

As months of supervision increases, the value of the

Probit increases, i.e. the greater the likelihood of
being successful on probation.

In Probit analysis, the equation of the best fit line is

achieved by maximum likelihood estimation.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 20

What is the Linear Relationship Between

Months of Supervision and Probation
Success?

The Probit model

Probit = a + b (months)

Estimated best fit linear equation

Probit = -0.70946 + 0.11582 (months)

Interpretation

For every one-month increase in the amount of

supervision, the Probit of success (Z score)
increases by 0.11582.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 21

Results of the Probit Analysis

Parameter estimates converged after 9 iterations.

Optimal solution found.

Parameter Estimates (PROBIT model: (PROBIT(p)) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTHS .11582 .03862 2.99914

Intercept Standard Error Intercept/S.E.

-.70946 .35511 -1.99788

Pearson Goodness-of-Fit Chi Square = .180 DF = 5 P = .999

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Probit Model

Probit = a + b (month)

a = intercept = -0.70946 (t = -1.99788, df = N - k

= 5, p< 0.05)

b = Probit regression coefficient = 0.11582

(t = 2.99914, df = N - k = 5, p< 0.05)

N = 7 months, k = 2 = number of parameters

estimated (a & b)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 22

Probit Model Predictions of the Effect of

Months of Supervision on Success

Probit = -0.70946 + 0.11582 (months)

Example: Predicted success of juveniles with 8

months supervision

Total = 36, 22 succeeded or 61.1%

Probit = -0.70946 + 0.11582 ( 8 months)

Probit = +0.2171

Z of +0.2171 has an associated area under

the larger portion of the curve of 0.5859
(36 juveniles) (0.5859) = 21.09 successes

Residual = (22 - 21.09) = 0.91

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 23

Probit Transformation

22 out of 36 juveniles succeeded with 8 months of

supervision (61.1%). The Probit equation yields a Z
= +0.2171. Since Z is positive, the area in the larger
portion of the curve is 0.5859, or a prediction of a
58.59 % success rate.

Standard Normal Curve

Area = 0.5859

Z = 0.00
Z = + 0.2171

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 24

Scatterplot of the Residuals Produced By

the Probit Model

Probit Residuals

0.00

1 2 3 4 5 6 7 8 9 10

Months of Supervision

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 25

Chi Square Goodness of Fit Test

λ 2 = Σ { (residual)2 / [ ni Pi (1-Pi) ] }

Residual = (observed – expected successes)

nI = Number of cases in the group, i.e. level of IV

PI = Predicted proportion of cases (cf. the next

exhibit for a summary of the predictions and the
residuals)

λ 2 = (-0.78)2 / [(38)(0.4942)(1-0.4942)] +
(0.36)2 / [(42)(0.54034)(1-0.54034)] + … +
(-0.075)2 / [(40)(0.75187)(1-0.75187)] = 0.18

λ 2
= 0.18, df = (g-k) = (7-2) = 5, p = 0.999

g = number of supervision groups, k = number of

parameters estimated (i.e. 2, a & b)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 26

Decision: Accept H0, no difference between the

observed and predicted proportion of successes.
The Probit model fits the data.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 27

Results of the Chi-Square Analysis of the

Residuals

Observed and Expected Frequencies

Number of Observed Expected

MONTHS Subjects Responses Responses Residual Prob

6.00 38.0 18.0 18.780 -.780 .49420

7.00 42.0 23.0 22.694 .306 .54034
8.00 36.0 22.0 21.094 .906 .58593
9.00 30.0 19.0 18.912 .088 .63040
10.00 48.0 32.0 32.313 -.313 .67319
11.00 45.0 32.0 32.122 -.122 .71381
12.00 40.0 30.0 30.075 -.075 .75187

Pearson Goodness-of-Fit Chi Square = .180 DF = 5 P = .999

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Prob = expected probability of success

Prob = (expected responses / number of subjects)

The null hypothesis that there is no significant

difference between the observed and expected
number of successful probationers is accepted.

Decision The Probit model fits the data.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 28

Table of Probit Model Predictions

Confidence Limits for Effective MONTHS

95% Confidence Limits

Prob MONTHS Lower Upper

.01 -13.96029 -57.27643 -4.83864

.02 -11.60665 -50.48995 -3.40871
.03 -10.11335 -46.18498 -2.50064
.04 -8.98999 -42.94706 -1.81699
.05 -8.07623 -40.31367 -1.26048
.06 -7.29847 -38.07258 -.78647
.07 -6.61653 -36.10789 -.37055
.08 -6.00593 -34.34901 .00213
.09 -5.45062 -32.74962 .34131
.10 -4.93945 -31.27762 .65376
.15 -2.82309 -25.18616 1.95040
.20 -1.14107 -20.34978 2.98586
.25 .30196 -16.20613 3.87972
.30 1.59784 -12.49186 4.68929
.35 2.79866 -9.05938 5.44882
.40 3.93813 -5.81619 6.18344
.45 5.04057 -2.70143 6.91725
.50 6.12554 .32014 7.68324
.55 7.21051 3.24192 8.54902
.60 8.31296 5.93250 9.70702
.65 9.45242 7.99500 11.62233
.70 10.65325 9.32416 14.48518
.75 11.94913 10.35800 17.97518
.80 13.39215 11.35755 22.01314
.85 15.07417 12.45162 26.79091
.90 17.19054 13.78504 32.84560
.91 17.70171 14.10331 34.31178
.92 18.25702 14.44795 35.90570
.93 18.86762 14.82578 37.65943
.94 19.54956 15.24661 39.61921
.95 20.32731 15.72535 41.85558
.96 21.24108 16.28646 44.48435
.97 22.36443 16.97471 47.71768
.98 23.85774 17.88757 52.01787
.99 26.21138 19.32293 58.79891

Median potency
Number of months of supervision which yeilds a Probit Z score =
0.0, 50% success rate, 6.12554 months
Probit = -0.70946 + 0.11582 (month)
0.0 = -0.70946 + 0.11582 (month)
month = (0.70946 / 0.11582) = 6.12554

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 29

Model 2
Logarithmic Transformed Probit Model of
the Effect of Months of Supervision on
Probation Success

Probit = a + b (log months)

Probit = -1.85282 + 2.31804 (log months)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 30

Logarithmic Transformation of Probit

Probit assumes a linear relationship between the IV

and the Z score transformation of the DV.

Sometimes the goodness of fit of a Probit model can

be improved by a logarithmic transformation of the
independent variable (s), e.g. months.

Model 2: Log transformed Probit model

A common logarithmic transformation is used,

i.e. base 10.

Probit = a + b ( log months )

Probit = -1.85282 + 2.31804 ( log months )

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 31

What is a Logarithm?

The logarithm of a number (N) is the exponent (x) to

which a base number (b) must be raised to yield that
number (N).

Log b N = x (b) X = N

Common logarithms

Base 10 (symbolized log)

Log 6 = 0.77815

(10)0.77815 = 5.9999 ≅ 6.0

Natural logarithms

Base e, e = 2.71828 (symbolized ln)

ln 6 = 1.791759

(e)1.791759 = 5.9999 ≅ 6.0

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 32

The Base of the Natural Logarithms (e)

The base of the natural logarithms is defined as the

limit of the following function (e = 2.71828):

e = lim (1 + 1/u)u
u➙∞

U (1 + 1/u)u

1 2.00000
10 2.59374
100 2.70481
1000 2.71692
10,000 2.71815
100,000 2.71827
1,000,000 2.71828 = e

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 33

Algebraic Laws of Logarithms

The example uses common logarithms, but the

same laws apply to natural logarithms.

1st Law: law of multiplication

log (3 x 2) = log 3 + log 2

log (6) = (90.47712 + 0.30103)

0.77815 = 0.77815

2nd Law: law of division

log (3 / 2) = log 3 - log 2

log 1.5 = (0.47712 - 0.30103)

0.17609 = 0.17609

3rd Law: law of exponents

log (3)2 = 2 (log 3)

log (9) = 2 (0.47712) = 0.95424

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 34

Logarithmic Transformation of
Months of Supervision

The objective of the transformation is to linearize the

relationship between months of supervision and the
proportion of successes.

Two options
Common logarithmic transformation (log)
Natural logarithmic transformation (ln)

Months Log ln
Base 10 Base e

6 0.778 1.792
7 0.845 1.946
8 0.903 2.079
9 0.954 2.197
10 1.000 2.303
11 1.041 2.398

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 35

12 1.079 2.485
Plot of the Z Transformed Proportion of
Successful Juvenile Probationers as a
Function of the Log 10 of Months of
Supervision

Probit Transformed Responses

.8

.6

.4

.2

0.0

-.2
.7 .8 .9 1.0 1.1

Log of MONTHS

Compare this with the plot of the Probit as a function

of months of supervision in Model 1. (cf. p. 17)

Is the relationship more graphically linear as a result

of the transformation of the number of months of
supervision to the log of months?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 36

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 37

Results of the Log 10 Transfomed

Probit Analysis
Parameter estimates converged after 10 iterations.
Optimal solution found.

Parameter Estimates (PROBIT model: (PROBIT(p)) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTHS 2.31804 .76881 3.01510

Intercept Standard Error Intercept/S.E.

-1.85282 .72824 -2.54425

Pearson Goodness-of-Fit Chi Square = .105 DF = 5 P = 1.000

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Probit = -1.85282 + 2.31804 (log month)

Prediction of success after 8 months

Probit = -1.85282 + 2.31804 (log 8)

Probit = -1.85282 + 2.31804 (0.903089)

= +0.240579

A Z of +0.240579 has a larger area of 0.59506

Therefore the predicted number of successes equals

(36 total cases) (0.59506 successes) = 21.402
successful cases

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 38

Analysis of the Residuals

Observed and Expected Frequencies

Number of Observed Expected

MONTHS Subjects Responses Responses Residual Prob

.78 38.0 18.0 18.257 -.257 .48045

.85 42.0 23.0 22.775 .225 .54227
.90 36.0 22.0 21.422 .578 .59506
.95 30.0 19.0 19.208 -.208 .64026
1.00 48.0 32.0 32.597 -.597 .67911
1.04 45.0 32.0 32.070 -.070 .71266
1.08 40.0 30.0 29.670 .330 .74175

Pearson Goodness-of-Fit Chi Square = .105 DF = 5 P = 1.000

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Test of goodness of fit

Chi-square = 0.105, p = 1.00

There is no significant difference between the

actual and predicted number of successes.

Decision The model fits the data.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 39

Table of Probit Model Predictions

Confidence Limits for Effective MONTHS

95% Confidence Limits

Prob MONTHS Lower Upper

.01 .62477 .00461 1.77880

.02 .81907 .00999 2.09760
.03 .97259 .01631 2.32911
.04 1.10678 .02358 2.52012
.05 1.22946 .03183 2.68711
.06 1.34455 .04109 2.83805
.07 1.45429 .05140 2.97746
.08 1.56012 .06280 3.10818
.09 1.66304 .07535 3.23213
.10 1.76378 .08910 3.35068
.15 2.25003 .17833 3.89097
.20 2.73042 .30936 4.38435
.25 3.22352 .49595 4.86029
.30 3.74178 .75713 5.33580
.35 4.29613 1.11933 5.82412
.40 4.89791 1.61946 6.33891
.45 5.56025 2.30899 6.89853
.50 6.29948 3.25719 7.53533
.55 7.13698 4.54260 8.32544
.60 8.10211 6.17111 9.50913
.65 9.23701 7.81138 11.82974
.70 10.60548 9.10228 16.38213
.75 12.31059 10.25491 24.37005
.80 14.53380 11.50799 38.59418
.85 17.63684 13.05549 66.50014
.90 22.49907 15.22534 132.52929
.91 23.86189 15.79445 156.61674
.92 25.43613 16.43471 187.79510
.93 27.28722 17.16649 229.31689
.94 29.51429 18.01996 286.66982
.95 32.27697 19.04257 369.83707
.96 35.85492 20.31525 498.94425
.97 40.80159 21.99299 721.10866
.98 48.44956 24.43386 1176.86871
.99 63.51682 28.83116 2547.90778

Median potency

50% success rate is predicted at 6.29948 months of

supervision

Model 1 predicted a median potency at 6.29948

months

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 40

Comparison of Models 1 & 2

The Effect of Log10 Transformation of
Months of Supervision

Transformation
Statistic Model 1 Model 2
None Log10

a (the constant) -0.70946 -1.85282

ta -1.99788 -2.54425
b (Probit coefficient) +0.11582 +2.31804
tb +2.99914 +3.01510
λ 2 (goodness of fit) 0.180 0.105
p of λ 2 (significance) 0.999 1.000
Σ (residuals)2 1.649 0.964
Median Potency (MP)
(in months) 6.126 6.299

The log transform model produced a slightly more

significant fit and slightly lower residuals. However
the operational difference between the two models is
negligible.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 41

Prediction With the Log Transformed

Probit Model

Probit = -1.85282 + 2.31804 (log months)

Example:: Expected success after 8 months of supervision.

Probit = -1.85282 + 2.31804 (log 8)

Probit = -1.85282 + 2.31804 (0.903)
Probit = +0.2406
Z score of +0.2406 divides a standard normal
curve into two parts corresponding to 0.5951 &
0.4049
area = 0.5951

area = 0.4049

Given 36 juveniles
(36) (0.5951) = 21.42
21 juveniles expected to
be successful

Z = +0.2406

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 42

Interpretation of the Regression Coefficient

(b) in the Probit Log Transformed Model

Probit = a + b (log months)

Probit = -1.85282 + 2.31804 (log months)

Interpretation: For every unit increase in the

log month

The Probit (Z score) increases

by 2.31804

Expected
Month Log Month Probit Successful

10 1.0 0.46522 0.681

100 2.0 2.78326 0.997

Difference = 2.31804

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 43

Model 3
Logit Model of the Effect of Months of
Supervision on Probation Success

Logit = a + b (months)

Logit = -1.15774 + 0.18841 (month)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 44

Logit: An Alternative Linear

Probability Model

Transform (Li) = a + bXI

Dependency Technique
One or more metric IV(s) reduced to fixed
levels or “doses”
One binary DV, measured as the proportion
of the subjects responding to a fixed level of
the IV. DV is transformed to ln [Pi / (1-Pi)].
Parameters are estimated by maximum
likelihood estimation.

Assumptions
That there is a linear relationship between
levels of the IV(s) and the DV
Homogeneity of variance of the DV over
levels of the IV
Homoskedastic residuals

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 45

Absence of outliers
The Logit Model

Transform (Li) = a + bXI

LI = Logit = ln [Pi / (1-Pi)],

ln = natural logarithm

Pi = the proportion of subjects in a group

effected by a level of the IV (e.g. proportion
successful)
a = constant
b = Logit regression coefficient
Xi = Level of the independent variable (e.g.
months of supervision)

Example: 22 out of 36 succeeded with 8 months

supervision
(22) / (36) = 0.611
Logit = ln [0.6111 / (1 - 0.6111)] = 0.447

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 46

Comparison of Probit and Logit

Transformations

Month Total Number P Probit Logit

Juveniles Success Success (Z score) ln [Pi / (1-Pi)]

6 38 18 0.47 -0.07 -0.120

7 42 23 0.55 +0.13 +0.201

8 36 22 0.61 +0.28 +0.447

9 30 19 0.63 +0.33 +0.532

10 48 32 0.66 +0.44 +0.663

11 45 32 0.71 +0.55 +0.895

12 40 30 0.75 +0.68 +1.099

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 47

Plot of the Logit as a Function of

Months of Supervision

Logit Transformed Responses

1.2

1.0

.8

.6

.4

.2

0.0

-.2
5 6 7 8 9 10 11 12 13

MONTHS

Notice how the logit transformation of the proportion

of successes has linearized the relationship between
success and months of supervision.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 48

Results of the Logit Analysis of Probation

Success as a Function of Months of
Supervision

Parameter estimates converged after 8 iterations.

Optimal solution found.

Parameter Estimates (LOGIT model: (LOG(p/(1-p))) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTHS .18841 .06326 2.97832

Intercept Standard Error Intercept/S.E.

-1.15774 .57656 -2.00802

Pearson Goodness-of-Fit Chi Square = .171 DF = 5 P = .999

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Logit = -1.15774 + 0.18841 (month)

Predicted success with 8 months supervision

Logit = ln (P / 1-P) = -1.15774 + 0.18841 (8)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 49

Results of the Logit Analysis of Probation Success as a Function of Months of

Supervision (cont.)

ln (P / 1-P) = 0.34954

Taking the antilog of both sides of the equation

(P / 1 - P) = e 0.34954 = 2.71828 0.34954 = 1.4184

(P / 1 - P) = 1.4184

P = 1.4184 (1 - P) = 1.4184 - 1.4184 P

0 = 1.4184 - 1.4184 P - P = 1.4184 - 2.4184 P

P = 1.4184 / 2.4184 = 0.5865

With 8 months of supervision, it is predicted that

0.5865 proportion, or 58.65%, of the juveniles will be
successful.

Given 36 juveniles supervised for 8 months, the

model predicts that 21.11 will be successful

(36 juveniles) (0.5865 successful) = 21.11

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 50

Residual Analysis

Observed and Expected Frequencies

Number of Observed Expected

MONTHS Subjects Responses Responses Residual Prob

6.00 38.0 18.0 18.741 -.741 .49318

7.00 42.0 23.0 22.688 .312 .54020
8.00 36.0 22.0 21.114 .886 .58651
9.00 30.0 19.0 18.940 .060 .63134
10.00 48.0 32.0 32.353 -.353 .67401
11.00 45.0 32.0 32.129 -.129 .71398
12.00 40.0 30.0 30.034 -.034 .75086

Pearson Goodness-of-Fit Chi Square = .171 DF = 5 P = .999

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

The chi-square analysis indicates that there is not a

significant difference between the actual and
predicted number of successes.

Decision The Logit model fits the data.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 51

Table of Logit Model Predictions

Confidence Limits for Effective MONTHS

95% Confidence Limits

Prob MONTHS Lower Upper

.01 -18.24389 -70.52373 -7.42514

.02 -14.51112 -59.61405 -5.16623
.03 -12.30468 -53.16658 -3.82971
.04 -10.72281 -48.54497 -2.87072
.05 -9.48290 -44.92304 -2.11845
.06 -8.45906 -41.93276 -1.49679
.07 -7.58414 -39.37783 -.96515
.08 -6.81804 -37.14104 -.49926
.09 -6.13490 -35.14680 -.08350
.10 -5.51705 -33.34345 .29283
.15 -3.06168 -26.18059 1.79213
.20 -1.21304 -20.79354 2.92679
.25 .31383 -16.35043 3.87024
.30 1.64769 -12.47658 4.70201
.35 2.85917 -8.96820 5.46755
.40 3.99271 -5.70029 6.19860
.45 5.07966 -2.59097 6.92386
.50 6.14472 .40982 7.68044
.55 7.20978 3.30559 8.54203
.60 8.29673 5.96602 9.71617
.65 9.43027 7.98983 11.69133
.70 10.64176 9.30236 14.65272
.75 11.97561 10.35044 18.31026
.80 13.50248 11.39645 22.65080
.85 15.35112 12.58812 27.98084
.90 17.80650 14.12312 35.10802
.91 18.42434 14.50507 36.90573
.92 19.10748 14.92610 38.89471
.93 19.87358 15.39694 41.12655
.94 20.74850 15.93328 43.67678
.95 21.77234 16.55944 46.66256
.96 23.01225 17.31607 50.28014
.97 24.59412 18.27936 54.89744
.98 26.80057 19.62028 61.34050
.99 30.53333 21.88407 72.24531

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 52

Model 4
Logarithic Transformed Logit Model of the
Effect of Months of Supervision on
Probation Success

Logit = a + b (log month)

Logit = -3.00611 + 3.75813 (log month)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 53

Plot of the Logits as a Function of the Log

Transformed Months of Supervistion

Logit Transformed Responses

1.2

1.0

.8

.6

.4

.2

0.0

-.2
.7 .8 .9 1.0 1.1

Log of MONTHS

Compare this scatterplot with the plot associated

with Model 3. (cf. p. 43)

Did the log transformation of months improve the

linearity of the relationship between the logit and
months of supervision?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 54

Results of the Logit Log Transformed

Analysis

Parameter estimates converged after 13 iterations.

Optimal solution found.

Parameter Estimates (LOGIT model: (LOG(p/(1-p))) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTHS 3.75813 1.25339 2.99836

Intercept Standard Error Intercept/S.E.

-3.00611 1.18264 -2.54186

Pearson Goodness-of-Fit Chi Square = .108 DF = 5 P = 1.000

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Logit = -3.00611 + 3.75813 (log month)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 55

Prediction with Logarithic Transformed

Logit Model

Logit = -3.00611 + 3.75813 (log month)

Example: Predicted success of juveniles with 8

months supervision Total = 36, of which 22 .

succeeded or 61.1%

Logit = ln [Pi / (1-Pi)]

ln [Pi / (1-Pi)] = -3.00611+ 3.75813 ( log 8 mos)
ln [Pi / (1-Pi)] = +0.3878
Take the antilog of both sides of the equation
Pi / (1-Pi) = e 0.3878 = (2.71828)0.3878 = 1.4737
P = 0.5957 proportion successful

(36 juveniles) (0.5957) = 21.45 successes

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 56

Residual Analysis

Observed and Expected Frequencies

Number of Observed Expected

MONTHS Subjects Responses Responses Residual Prob

.78 38.0 18.0 18.224 -.224 .47958

.85 42.0 23.0 22.779 .221 .54237
.90 36.0 22.0 21.447 .553 .59576
.95 30.0 19.0 19.232 -.232 .64108
1.00 48.0 32.0 32.622 -.622 .67962
1.04 45.0 32.0 32.063 -.063 .71250
1.08 40.0 30.0 29.628 .372 .74070

Pearson Goodness-of-Fit Chi Square = .108 DF = 5 P = 1.000

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

The chi-square analysis indicates that there is not a

significant difference between the actual and
predicted number of successes.

Decision The Logit model fits the data.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 57

Table of Logit Model Predictions

Confidence Limits for Effective MONTHS

95% Confidence Limits

Prob MONTHS Lower Upper

.01 .37773 .00101 1.31099

.02 .58119 .00351 1.70275
.03 .74979 .00733 1.98764
.04 .90001 .01241 2.22098
.05 1.03851 .01875 2.42304
.06 1.16880 .02636 2.60382
.07 1.29302 .03527 2.76908
.08 1.41258 .04551 2.92251
.09 1.52848 .05712 3.06659
.10 1.64148 .07016 3.20312
.15 2.17940 .15871 3.81016
.20 2.69785 .29324 4.34493
.25 3.21786 .48656 4.84630
.30 3.75352 .75659 5.33610
.35 4.31694 1.12847 5.83053
.40 4.92047 1.63760 6.34536
.45 5.57828 2.33380 6.90093
.50 6.30807 3.28497 7.53211
.55 7.13334 4.56867 8.32023
.60 8.08699 6.18747 9.52183
.65 9.21758 7.80398 11.92296
.70 10.60118 9.08367 16.69367
.75 12.36592 10.25675 25.31287
.80 14.74945 11.57796 41.50041
.85 18.25815 13.29108 76.17175
.90 24.24132 15.87592 171.59932
.91 26.03346 16.59364 210.61476
.92 28.16962 17.42242 264.19811
.93 30.77436 18.39837 340.71509
.94 34.04496 19.57686 455.63122
.95 38.31622 21.04849 640.31575
.96 44.21252 22.97513 967.04695
.97 53.07036 25.68524 1636.75927
.98 68.46545 29.99797 3411.08530
.99 105.34466 38.98445 11820.75520

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 58

Comparison of Models 1, 2, 3, & 4

Type of
Model Probit Logit
Model 1 Model2 Model 3 Model 4
Statistic (log) (log)

b 0.11582 2.31804 0.18841 3.75813

t 2.99914 3.0151 2.97832 2.998

λ 2 0.180 0.105 0.171 0.108

(p) 0.999 1.000 0.999 1.000

Σ (residuals)2 1.649 0.964 1.577 0.988

Median Potency 6.1255 6.2995 6.1447 6.3081

(in months)

All four models produce simmilar results. The log

transform models produce the smaller residuals.

The Probit log model produces the smallest sum of

squared residuals and is the best of the four models.

Operationally, there is negligible difference among the four

models

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 59

Model 5
A Multivariate Probit Model with Two
Independent Variables
*****

Success as a Function of Months of

Supervision and Age

Probit = a + b1 (month) + b2 (age)

Probit = 7.06114 + 0.13009 (month) – 0.56226 (age)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 60

Results of the Probit Analysis of Probation

Success as a Function of Months of
Supervision and Age

Parameter estimates converged after 14 iterations.

Optimal solution found.

Parameter Estimates (PROBIT model: (PROBIT(p)) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.
MONTH .13009 .04005 3.24841
AGE -.56226 .10113 -5.55984
Intercept Standard Error Intercept/S.E.
7.06114 1.43287 4.92797

Pearson Goodness-of-Fit Chi Square = 26.778 DF = 39 P = .931

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

Probit = 7.06114 + 0.13009 (month) – 0.56226 (age)

Interpretation

When supervision increases by 1 month, the probit of

success increases by +0.13009.

When age increases by one year, the probit of success

decreases by -0.56226.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 61

Predictions With the Probit Equation

Probit = 7.06114 + 0.13009 (month) – 0.56226 (age)

Example: Predicted success of 14 year olds with 8

months supervision

Total = 12, 8 succeeded or 66.67%

Probit = 7.06114 + 0.13009 (8 month) -

0.56226 (14 years)

Probit = +0.23022

Z of +0.23022 has an associated area under

the larger portion of the curve of 0.591
(12 juveniles) (0.591) = 7.09 successes

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 62

Covariate Analysis

To what extent are the two predictor variables

(covariates) collinear?

Correlation between age and months of

supervision: r = - 0.11185

Covariance between the predictor variables: cov

= - 0.00045

Covariance(below) and Correlation(above) Matrices of Parameter Estimates

MONTH AGE

MONTH .00160 -.11185

AGE -.00045 .01023

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 63

Table of Probit Model Predictions

Observed and Expected Frequencies

Number of Observed Expected

MONTH Subjects Responses Responses Residual Prob

6.00 7.0 3.0 4.919 -1.919 .70276

6.00 7.0 5.0 4.919 .081 .70276
6.00 6.0 2.0 2.928 -.928 .48807
6.00 6.0 4.0 2.928 1.072 .48807
6.00 6.0 2.0 1.661 .339 .27687
6.00 6.0 2.0 1.661 .339 .27687
7.00 7.0 4.0 5.223 -1.223 .74615
7.00 7.0 7.0 5.223 1.777 .74615
7.00 7.0 2.0 3.779 -1.779 .53990
7.00 7.0 5.0 3.779 1.221 .53990
7.00 7.0 2.0 2.254 -.254 .32201
7.00 7.0 3.0 2.254 .746 .32201
8.00 6.0 5.0 4.716 .284 .78597
8.00 6.0 6.0 4.716 1.284 .78597
8.00 6.0 3.0 3.546 -.546 .59106
8.00 6.0 5.0 3.546 1.454 .59106
8.00 6.0 1.0 2.220 -1.220 .36995
8.00 6.0 2.0 2.220 -.220 .36995
9.00 5.0 4.0 4.109 -.109 .82190
9.00 5.0 5.0 4.109 .891 .82190
9.00 5.0 3.0 3.204 -.204 .64071
9.00 5.0 3.0 3.204 -.204 .64071
9.00 5.0 2.0 2.100 -.100 .42000
9.00 5.0 2.0 2.100 -.100 .42000
10.00 8.0 6.0 6.830 -.830 .85376
10.00 8.0 8.0 6.830 1.170 .85376
10.00 8.0 5.0 5.505 -.505 .68809
10.00 8.0 6.0 5.505 .495 .68809
10.00 8.0 2.0 3.771 -1.771 .47138
10.00 8.0 5.0 3.771 1.229 .47138
11.00 8.0 6.0 7.052 -1.052 .88156
11.00 8.0 7.0 7.052 -.052 .88156
11.00 8.0 5.0 5.860 -.860 .73255
11.00 8.0 6.0 5.860 .140 .73255
11.00 7.0 3.0 3.663 -.663 .52324
11.00 7.0 5.0 3.663 1.337 .52324
12.00 6.0 5.0 5.432 -.432 .90539
12.00 6.0 6.0 5.432 .568 .90539
12.00 6.0 4.0 4.641 -.641 .77357
12.00 6.0 5.0 4.641 .359 .77357
12.00 8.0 4.0 4.598 -.598 .57471
12.00 8.0 6.0 4.598 1.402 .57471

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 64

Model 6
Probit Analysis With Multiple Groups
*****
Independent Variable = months of supervision
Groups = gender, males & females

Probit male = a1 + b1 (month)

Probit female = a2 + b2 (month)

Probit male = -1.001139 + 0.11726 (month)

Probit female = -0.41904 + 0.11726 (month)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 65

Probit Analysis With Multiple Groups

Sometimes it is of interest to compare the Probit

equations for multiple groups to determine if there
are significant differences in the Probit coefficients
(i.e. slopes) predictor variables across groups.

Examples of different groups

Racial/ethnic groups

Offense groups

Gender groups

Q Are there significant differences between boys

and girls in the effect of the number of months of
supervision on probation success?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 66

The Concept of Testing the Parallelism of

the Slopes of Multiple Group Probit
Equations

Step 1

Estimate separate Probit equations for boys and

girls on the effect of months of supervision on
probation success.

Step 2

Conduct a chi-square test to determine whether

the slopes (coefficients) of the two equations
differ significantly, i.e.

Whether the probit coefficients of the two

equations differ significantly.

Step 3

If the slopes do not differ significantly, compute

a common slope to be used for each group

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 67

Plot of the Probits of Probation Success for

Boys and Girls

Probit Transformed Responses

1.5

1.0

.5

0.0

-.5
GENDER
Female

-1.0 Male
5 6 7 8 9 10 11 12 13

MONTH

Q Is the linear relationship between of months of

supervision on success the same (parallel) for male
and female juvenile probationers?

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 68

Results of the Probit Analysis for

Boys and Girls

Parameter estimates converged after 8 iterations.

Optimal solution found.

Parameter Estimates (PROBIT model: (PROBIT(p)) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTH .11726 .03922 2.98946

Intercept Standard Error Intercept/S.E. GENDER

-1.01139 .37197 -2.71905 1

-.41904 .36667 -1.14281 2

Pearson Goodness-of-Fit Chi Square = 42.756 DF = 39 P = .313

Parallelism Test Chi Square = .000 DF = 1 P = 1.000

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

The parallelism test found no significant difference

between the slopes of the two equations, so a
common slope was computed to be used in each
equation, i.e. bC = 0.11726.

Probit male = -1.001139 + 0.11726 (month)

Probit female = -0.41904 + 0.11726 (month)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 69

The Concept of a Common Slope

Probit male = -1.001139 + 0.11726 (month)

Probit female = -0.41904 + 0.11726 (month)

Probit Z Score

Female Slope = 0.117

M

0.00 ✚

Male Slope = 0.117

-0.42 ✚

-1.02 ✚

0 1 2 3 4 5 6 7 8 9 10

Months Of Supervision

Notice that the difference between the two groups is in the

intercepts: females = - 0.42, males = - 1.0. Males have a

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 70

lower potential for success than females, regardless of the

amount of supervision.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 71

Predictions from Probit Equations for Male

& Female Probationers with 8 Months
Supervision

Male Probationers
Total = 18, 9 successful or 50%
Probit = -1.01139 + 0.11726 (8 months)
Probit = -0.07331
Z of -0.07331 has an associated area under
the smaller portion of the curve of 0.4708
(18 juveniles) (0.4708) = 8.475 successes

Female Probationers
Total = 18, 13 successful or 72.2%
Probit = -0.41904 + 0.11726 (8 months)
Probit = +0.51904
Z of +0.51904 has an associated area under
the larger portion of the curve of 0.69817
(18 juveniles) (0.69817) = 12.567 successes

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 72

Residual Analysis
Observed and Expected Frequencies

Number of Observed Expected

GENDER MONTH Subjects Responses Responses Residual Prob

1 6.00 7.0 3.0 2.654 .346 .37911

1 6.00 6.0 2.0 2.275 -.275 .37911
1 6.00 6.0 2.0 2.275 -.275 .37911
1 7.00 7.0 4.0 2.971 1.029 .42443
1 7.00 7.0 2.0 2.971 -.971 .42443
1 7.00 7.0 2.0 2.971 -.971 .42443
1 8.00 6.0 5.0 2.825 2.175 .47078
1 8.00 6.0 3.0 2.825 .175 .47078
1 8.00 6.0 1.0 2.825 -1.825 .47078
1 9.00 5.0 4.0 2.588 1.412 .51753
1 9.00 5.0 3.0 2.588 .412 .51753
1 9.00 5.0 2.0 2.588 -.588 .51753
1 10.00 8.0 6.0 4.512 1.488 .56404
1 10.00 8.0 5.0 4.512 .488 .56404
1 10.00 8.0 2.0 4.512 -2.512 .56404
1 11.00 8.0 6.0 4.877 1.123 .60968
1 11.00 8.0 5.0 4.877 .123 .60968
1 11.00 7.0 3.0 4.268 -1.268 .60968
1 12.00 6.0 5.0 3.923 1.077 .65385
1 12.00 6.0 4.0 3.923 .077 .65385
1 12.00 8.0 4.0 5.231 -1.231 .65385

2 6.00 7.0 5.0 4.284 .716 .61200

2 6.00 6.0 4.0 3.672 .328 .61200
2 6.00 6.0 2.0 3.672 -1.672 .61200
2 7.00 7.0 7.0 4.593 2.407 .65608
2 7.00 7.0 5.0 4.593 .407 .65608
2 7.00 7.0 3.0 4.593 -1.593 .65608
2 8.00 6.0 6.0 4.189 1.811 .69814
2 8.00 6.0 5.0 4.189 .811 .69814
2 8.00 6.0 2.0 4.189 -2.189 .69814
2 9.00 5.0 5.0 3.689 1.311 .73771
2 9.00 5.0 3.0 3.689 -.689 .73771
2 9.00 5.0 2.0 3.689 -1.689 .73771
2 10.00 8.0 8.0 6.196 1.804 .77445
2 10.00 8.0 6.0 6.196 -.196 .77445
2 10.00 8.0 5.0 6.196 -1.196 .77445
2 11.00 8.0 7.0 6.465 .535 .80808
2 11.00 8.0 6.0 6.465 -.465 .80808
2 11.00 7.0 5.0 5.657 -.657 .80808
2 12.00 6.0 6.0 5.031 .969 .83845
2 12.00 6.0 5.0 5.031 -.031 .83845
2 12.00 8.0 6.0 6.708 -.708 .83845

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 73

Table of Probit Model Predictions

For Males
Confidence Limits for Effective MONTH

GENDER 1 1

95% Confidence Limits

Prob MONTH Lower Upper

.01 -11.21390 -49.97444 -3.05681

.02 -8.88918 -43.23911 -1.63725
.03 -7.41422 -38.96799 -.73435
.04 -6.30467 -35.75649 -.05363
.05 -5.40213 -33.14536 .50126
.06 -4.63393 -30.92388 .97455
.07 -3.96037 -28.97696 1.39042
.08 -3.35727 -27.23454 1.76359
.09 -2.80878 -25.65064 2.10375
.10 -2.30390 -24.19340 2.41761
.15 -.21354 -18.17002 3.72702
.20 1.44782 -13.40033 4.78521
.25 2.87311 -9.33024 5.71491
.30 4.15307 -5.70589 6.58054
.35 5.33914 -2.39556 7.43084
.40 6.46461 .66083 8.32248
.45 7.55351 3.45078 9.35229
.50 8.62515 5.85315 10.70912
.55 9.69678 7.68676 12.63471
.60 10.78569 8.99479 15.14645
.65 11.91115 10.01863 18.07065
.70 13.09722 10.93403 21.31588
.75 14.37718 11.83491 24.90497
.80 15.80247 12.78553 28.95414
.85 17.46383 13.85715 33.71040
.90 19.55419 15.17585 39.72450
.91 20.05908 15.49124 41.18020
.92 20.60757 15.83286 42.76263
.93 21.21066 16.20743 44.50366
.94 21.88422 16.62464 46.44924
.95 22.65242 17.09924 48.66941
.96 23.55496 17.65542 51.27925
.97 24.66452 18.33744 54.48944
.98 26.13947 19.24171 58.75919
.99 28.46419 20.66285 65.49295

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 74

Table of Probit Model Predictions (cont.)

For Females
Confidence Limits for Effective MONTH

GENDER 2 2

95% Confidence Limits

Prob MONTH Lower Upper

.01 -16.26550 -64.24189 -6.19771

.02 -13.94078 -57.50252 -4.78220
.03 -12.46582 -53.22787 -3.88282
.04 -11.35626 -50.01304 -3.20544
.05 -10.45373 -47.39862 -2.65384
.06 -9.68553 -45.17383 -2.18386
.07 -9.01197 -43.22355 -1.77135
.08 -8.40887 -41.47767 -1.40163
.09 -7.86038 -39.89020 -1.06504
.10 -7.35550 -38.42925 -.75491
.15 -5.26513 -32.38438 .53303
.20 -3.60378 -27.58605 1.56257
.25 -2.17848 -23.47560 2.45192
.30 -.89853 -19.79114 3.25743
.35 .28754 -16.38519 4.01211
.40 1.41301 -13.16393 4.73888
.45 2.50191 -10.06199 5.45670
.50 3.57355 -7.03097 6.18489
.55 4.64519 -4.03520 6.94832
.60 5.73409 -1.05482 7.78771
.65 6.85955 1.89443 8.78649
.70 8.04563 4.69772 10.14383
.75 9.32558 7.06270 12.26882
.80 10.75088 8.82453 15.50679
.85 12.41223 10.27770 19.88150
.90 14.50259 11.78928 25.70271
.91 15.00748 12.13165 27.13144
.92 15.55597 12.49769 28.68944
.93 16.15907 12.89456 30.40817
.94 16.83263 13.33230 32.33323
.95 17.60083 13.82601 34.53429
.96 18.50336 14.40025 37.12606
.97 19.61292 15.09973 40.31880
.98 21.08788 16.02155 44.57100
.99 23.41260 17.46195 51.28549

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 75

Model 7
Multivariate Probit Models with Multiple
Groups and Two Predictor Variables

Groups = males & females

Predictors = month & age

Probit male = a + b1 (month) + b2 (age)

Probit female = a + b1 (month) + b2 (age)

Probit male = 7.17239 + 0.13603 (month) – 0.59618 (age)

Probit female = 7.83981 + 0.13603 (month) – 0.59618 (age)

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 76

Results of the Probit Analysis

Parameter estimates converged after 17 iterations.
Optimal solution found.

Parameter Estimates (PROBIT model: (PROBIT(p)) = Intercept + BX):

Regression Coeff. Standard Error Coeff./S.E.

MONTH .13603 .04103 3.31503

AGE -.59618 .10452 -5.70371

Intercept Standard Error Intercept/S.E. GENDER

7.17239 1.47404 4.86581 1

7.83981 1.49292 5.25132 2

Pearson Goodness-of-Fit Chi Square = 12.938 DF = 38 P = 1.000

Parallelism Test Chi Square = .139 DF = 1 P = .709

Since Goodness-of-Fit Chi square is NOT significant, no heterogeneity

factor is used in the calculation of confidence limits.

The paraellelism chi-square test found no significant

differences between the group slopes for month and
age and computed a common slope for each
variable for each group.

Probit male = 7.17239 + 0.13603 (month) – 0.59618 (age)

Probit female = 7.83981 + 0.13603 (month) – 0.59618 (age)

The groups differ in the intercept; males have a

lower likelihood of success regardless of months of
supervision or age.

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 77

Residual Analysis
Observed and Expected Frequencies

Number of Observed Expected

GENDER MONTH Subjects Responses Responses Residual Prob

1 6.00 7.0 3.0 4.159 -1.159 .59417

1 6.00 6.0 2.0 2.161 -.161 .36021
1 6.00 6.0 2.0 1.020 .980 .17002
1 7.00 7.0 4.0 4.521 -.521 .64591
1 7.00 7.0 2.0 2.885 -.885 .41221
1 7.00 7.0 2.0 1.447 .553 .20666
1 8.00 6.0 5.0 4.171 .829 .69509
1 8.00 6.0 3.0 2.795 .205 .46580
1 8.00 6.0 1.0 1.486 -.486 .24761
1 9.00 5.0 4.0 3.705 .295 .74098
1 9.00 5.0 3.0 2.600 .400 .52002
1 9.00 5.0 2.0 1.463 .537 .29254
1 10.00 8.0 6.0 6.264 -.264 .78301
1 10.00 8.0 5.0 4.591 .409 .57386
1 10.00 8.0 2.0 2.727 -.727 .34092
1 11.00 8.0 6.0 6.566 -.566 .82080
1 11.00 8.0 5.0 5.011 -.011 .62637
1 11.00 7.0 3.0 2.745 .255 .39207
1 12.00 6.0 5.0 5.125 -.125 .85416
1 12.00 6.0 4.0 4.060 -.060 .67663
1 12.00 8.0 4.0 3.561 .439 .44516

2 6.00 7.0 5.0 5.722 -.722 .81745

2 6.00 6.0 4.0 3.729 .271 .62154
2 6.00 6.0 2.0 2.323 -.323 .38719
2 7.00 7.0 7.0 5.959 1.041 .85123
2 7.00 7.0 5.0 4.704 .296 .67204
2 7.00 7.0 3.0 3.081 -.081 .44014
2 8.00 6.0 6.0 5.283 .717 .88055
2 8.00 6.0 5.0 4.317 .683 .71958
2 8.00 6.0 2.0 2.965 -.965 .49418
2 9.00 5.0 5.0 4.528 .472 .90554
2 9.00 5.0 3.0 3.818 -.818 .76350
2 9.00 5.0 2.0 2.742 -.742 .54833
2 10.00 8.0 8.0 7.412 .588 .92645
2 10.00 8.0 6.0 6.427 -.427 .80335
2 10.00 8.0 5.0 4.813 .187 .60159
2 11.00 8.0 7.0 7.549 -.549 .94361
2 11.00 8.0 6.0 6.711 -.711 .83883
2 11.00 7.0 5.0 4.571 .429 .65302
2 12.00 6.0 6.0 5.745 .255 .95745
2 12.00 6.0 5.0 5.219 -.219 .86986
2 12.00 8.0 6.0 5.614 .386 .70178

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 78

Predictions from Probit Equations for 14

Year Old Male & Female Probationers with
8 Months of Supervision

Male Probationers
Total = 6, 3 successful or 50%
Probit = 7.17239 + 0.13603 (8 months) -
0.59618 (14 years old)
Probit = -0.08589
Z of -0.08589 has an associated area under
the smaller portion of the curve of 0.4658
(6 juveniles) (0.4658) = 2.795 successes

Female Probationers
Total = 6, 5 successful or 83.3%
Probit = 7.83981 + 0.13603 (8 months) -
0.59618 (14 years old)
Probit = +0.58153
Z of +0.58153 has an associated area under
the larger portion of the curve of 0.71958
(6 juveniles) (0.71958) = 4.317 successes

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University
 79

Probit/Logit Analysis: Charles M. Friel Ph.D., Criminal Justice Center, Sam Houston State University