Вы находитесь на странице: 1из 8

 Introduction

A binomial logistic regression (often referred to simply as logistic regression),


predicts the probability that an observation falls into one of two categories of a
dichotomous dependent variable based on one or more independent variables that can
be either continuous or categorical. If, on the other hand, your dependent variable is a
count, see our Poisson regression guide. Alternatively, if you have more than two
categories of the dependent variable, see our multinomial logistic regression guide.
For example, you could use binomial logistic regression to understand whether exam
performance can be predicted based on revision time, test anxiety and lecture
attendance (i.e., where the dependent variable is "exam performance", measured on a
dichotomous scale – "passed" or "failed" – and you have three independent variables:
"revision time", "test anxiety" and "lecture attendance"). Alternately, you could use
binomial logistic regression to understand whether drug use can be predicted based on
prior criminal convictions, drug use amongst friends, income, age and gender (i.e.,
where the dependent variable is "drug use", measured on a dichotomous scale – "yes"
or "no" – and you have five independent variables: "prior criminal convictions", "drug
use amongst friends", "income", "age" and "gender").
 The steps for interpreting the SPSS output for a logistic regression
1. Scroll down to the Block 1: Method = Enter section of the output.
2. Look in the Omnibus Tests of Model Coefficients table, under
the Sig. column, in the Model row. This is the p-value that is interpreted.
 If the p-value is LESS THAN .05, then researchers have a significant
model that should be further interpreted.
If the p-value is MORE THAN .05, then researchers do not have a significant
model and the results should be reported.
3. Look in the Hosmer and Lemeshow Test table, under the Sig. column. This is
the p-value you will interpret.
 If the p-value is LESS THAN .05, then the model does not fit the data.
 If the p-value is MORE THAN .05, then the model does fit the data and
should be further interpreted.
4. Look in the Classification Table, under the Percentage Correct in the Overall
Percentage row. This is the total accuracy of the model. Researchers want it to
ultimately be at least 80%.
5. Look in the Variables in the Equation table, under the Sig., Exp(B),
and Lower and Upper columns. The Sig. column is the p-value associated with
the adjusted odds ratios and 95% CIs for each predictor, clinical, demographic, or
confounding variable. The value in the Exp(B) is the adjusted odds ratio.
The Lower and Upper values are the limits of the 95% CI associated with the
adjusted odds ratio. 
6. Researchers will interpret the adjusted odds ratio in the Exp(B) column and the
confidence interval in the Lower and Upper columns for each variable.
 If the confidence interval associated with the adjusted ratio crosses over 1.0,
then there is a non-significant association. The p-value associated with
these variables will also be HIGHER than .05. 
 If the adjusted odds ratio is ABOVE 1.0 and the confidence interval is
entirely above 1.0, then exposure to the predictor increases the odds of the
outcome.
 If the adjusted odds ratio is BELOW 1.0 and the confidence interval is
entirely below 1.0, then exposure to the predictor decreases the odds of the
outcome.
 If the variable is measured at the ordinal or continuous level, then the
adjusted odds ratio is interpreted as meaning for every one unit increase in
the ordinal or continuous variable, the risk of the outcome increases at the
rate specified in the odds ratio.   

 Example 1
A health researcher wants to be able to predict whether the "incidence of heart
disease" can be predicted based on "age", "weight", "gender" and "VO2max" (i.e.,
where VO2max refers to maximal aerobic capacity, an indicator of fitness and health).
To this end, the researcher recruited 100 participants to perform a maximum VO2max
test as well as recording their age, weight and gender. The participants were also
evaluated for the presence of heart disease. A binomial logistic regression was then
run to determine whether the presence of heart disease could be predicted from their
VO2max, age, weight and gender.
Interpreting and Reporting the Output of a Binomial Logistic Regression Analysis
SPSS Statistics generates many tables of output when carrying out binomial logistic
regression. In this section, we show you only the three main tables required to
understand your results from the binomial logistic regression procedure, assuming that
no assumptions have been violated.
 Variance explained
In order to understand how much variation in the dependent variable can be explained
by the model (the equivalent of R2 in multiple regression), you can consult the table
below, "Model Summary":
This table contains the Cox & Snell R Square and Nagelkerke R Square values, which
are both methods of calculating the explained variation. These values are sometimes
referred to as pseudo R2 values (and will have lower values than in multiple
regression). However, they are interpreted in the same manner, but with more caution.
Therefore, the explained variation in the dependent variable based on our model
ranges from 24.0% to 33.0%, depending on whether you reference the Cox & Snell R 2
or Nagelkerke R2 methods, respectively. Nagelkerke R 2 is a modification of Cox &
Snell R2, the latter of which cannot achieve a value of 1. For this reason, it is
preferable to report the Nagelkerke R2 value.
 Category prediction
Binomial logistic regression estimates the probability of an event (in this case, having
heart disease) occurring. If the estimated probability of the event occurring is greater
than or equal to 0.5 (better than even chance), SPSS Statistics classifies the event as
occurring (e.g., heart disease being present). If the probability is less than 0.5, SPSS
Statistics classifies the event as not occurring (e.g., no heart disease). It is very
common to use binomial logistic regression to predict whether cases can be correctly
classified (i.e., predicted) from the independent variables. Therefore, it becomes
necessary to have a method to assess the effectiveness of the predicted classification
against the actual classification. There are many methods to assess this with their
usefulness often depending on the nature of the study conducted. However, all
methods revolve around the observed and predicted classifications, which are
presented in the "Classification Table", as shown below:

Firstly, notice that the table has a subscript which states, "The cut value is .500". This
means that if the probability of a case being classified into the "yes" category is
greater than .500, then that particular case is classified into the "yes" category.
Otherwise, the case is classified as in the "no" category (as mentioned previously).
Whilst the classification table appears to be very simple, it actually provides a lot of
important information about your binomial logistic regression result, including:
 The percentage accuracy in classification (PAC), which reflects the percentage
of cases that can be correctly classified as "no" heart disease with the
independent variables added (not just the overall model).
 Sensitivity, which is the percentage of cases that had the observed
characteristic (e.g., "yes" for heart disease) which were correctly predicted by
the model (i.e., true positives).
 Specificity, which is the percentage of cases that did not have the observed
characteristic (e.g., "no" for heart disease) and were also correctly predicted as
not having the observed characteristic (i.e., true negatives).
 The positive predictive value, which is the percentage of correctly predicted
cases "with" the observed characteristic compared to the total number of cases
predicted as having the characteristic.
 The negative predictive value, which is the percentage of correctly predicted
cases "without" the observed characteristic compared to the total number of
cases predicted as not having the characteristic.
 Variables in the equation
The "Variables in the Equation" table shows the contribution of each independent
variable to the model and its statistical significance. This table is shown below:

The Wald test ("Wald" column) is used to determine statistical significance for each of
the independent variables. The statistical significance of the test is found in the "Sig."
column. From these results you can see that age (p = .003), gender (p = .021) and
VO2max (p = .039) added significantly to the model/prediction, but weight (p = .799)
did not add significantly to the model. You can use the information in the "Variables
in the Equation" table to predict the probability of an event occurring based on a one
unit change in an independent variable when all other independent variables are kept
constant. For example, the table shows that the odds of having heart disease ("yes"
category) is 7.026 times greater for males as opposed to females. If you are unsure
how to use odds ratios to make predictions, learn about our enhanced guides here.
Based on the results above, we could report the results of the study as follows:
A logistic regression was performed to ascertain the effects of age, weight, gender and
VO2max on the likelihood that participants have heart disease. The logistic regression
model was statistically significant, χ2(4) = 27.402, p < .0005. The model explained
33.0% (Nagelkerke R2) of the variance in heart disease and correctly classified 71.0%
of cases. Males were 7.02 times more likely to exhibit heart disease than females.
Increasing age was associated with an increased likelihood of exhibiting heart disease,
but increasing VO2max was associated with a reduction in the likelihood of exhibiting
heart disease.

Example 2: a dataset of college student political behaviors and attitudes. In the


example, I am trying see if whether or not a student said that he voted (coded 1="yes";
0="no") can be predicted by the following four variables:
 Gender (Variable "q1," a dichotomous variable where male respondents are
coded "1" and women="0")
 Political ideology (Variable "q16," which is coded in discreet, one unit
intervals where 1="very conservative" and 7="very liberal."
 Cumulative GPA (Variable "q9," coded as a continuous variable that can range
from o.0-4.0).
 Follow politics closely (Variable "q17L," which is coded in one unit, discreet
intervals ranging from 1-5. Students's answers to the question, "I follow politics
closely" in this survey ranged from 1 ("very much disagree") to 5 ("very much
agree").

Interpretation:
The results in the SPSS output window will have many tables; we are interested only
in the following two tables (pay close attention to the table that are listed towards the
very end of the output):

Although the logic and method of calculation used in logistic regression is different
than that used for regular regression, SPSS provides two "pseudo R-squared statistics"
(this is the term we use when we report this data), that can be interpreted in a way that
is similar to that in multiple regression. The main difference between the Cox and
Snell measurement and the Nagelkerke measure is that the former tends to produce
more conservative (that is lower) pseudo R2s than the latter measure.
In political science, most researchers use the more conservative Cox and Snell pseudo
R2 statistic. The Cox and Snell pseudo R2 statistic reported in Figure 3 is generally
interpreted to mean:
"the four independent variables in the logistic model together account for 15.7 percent
the explanation for why a student votes or not."
Generally speaking, the higher the pseudo R-squared statistic, the better the model fits
our data. In this case, we would probably say that the model we have built
"moderately" fits our data (in other words, although the model accounts for a
significant amount of the variation in whether or not a student votes, there are also
lots of other variables not in our model which influence this decision).
You should be aware of the fact that there is much debate among scholars over which
statistics should be reported when using logistic regression, and many articles and
books using this technique will employ other measures to assess how well a given
logistic regression model "fits"--that is precisely includes the correct independent
variables and only the right variables. Nevertheless, the reason the Cox and Snell
pseudo R-squared statistic is automatically calculated by SPSS is because it is both
widely reported and fairly straightforward to understand and explain. It closely
resembles the much more universally accepted R-squared statistic that we use to
assess model fit when using OLS multiple regression.

Interpreting how much each of independent variable contributes to variations in the


dependent variable when controlling for other variables. Figure 4 above reports the
partial logistic regression coefficients for each independent variable in the model in
the column marked "B." PLEASE NOTE: THESE COEFFICIENTS DO NOT HAVE
THE SAME MEANING IN LOGISTIC REGRESSION THAT THEY HAVE IN
REGULAR BIVARIATE AND MULTIPLE REGRESSION!!! By themselves, these
coefficients do not have a meaning that is easily explained or understood except by
experts.
To assess the isolated impact of each independent variable, we instead want look at
what are called the "odds ratios," which are listed in Figure 4 in the column titled
Exp(B).
The easiest way to explain how to interpret an odds ratio is to use an example from the
table. Recall that Q1 is the variable name for the gender variable (male=1; female=0).
What the logistic regression results in Figure 4's Exp(B) collumn say is:
"Controlling for differences in political ideology, GPA, and the extent to one agrees
with the statement 'I follow politics closely,' being a male student increases the
likelihood of voting by 1.46 times."
The political ideology variable is coded 1-7, where each one unit increase means that
a student self-identified as being increasingly "liberal." Interpreting odds-ratio for Q16
(the variable name for the ideology variable) indicates that:
"For every one unit increase in a student's liberalism (as measured by a 7-unit index),
the likelihood of voting decreased slightly (by .96 times), after controlling for the
other factors in the model."
The other two independent variables might be interpreted in the following manner:
"For each full grade increase in cumulative GPA (one full point on the four point
grading scale), students were nearly four times as likely to vote, controlling for all
factors included in the model. In interpreting this figure, it is important to keep in
mind that the great majority of students in this sample have grade point averages that
range between 3.0 and 3.9."
"Students who report that they follow politics regularly were much more likely to
vote. Students were coded on a five point discreet scale ranging from strongly
disagree (1) to strongly agree (5) with the statement 'I follow politics closely.' In the
logistic regression model, every one-unit shift towards the "strongly agree" category
corresponded with an increased likelihood of voting by 1.87 times."
As with the pseudo R-square statistic, there is some debate over how logistic partial
regression statistics should be interpreted, which means that you may read logistic
regression tables where other measures are used. Unfortunately, not all social
scientists using logistic regression will report odds-ratios. SPSS reports this statistic
because they it is a widely-used and easily-understood measure of how each the
independent variable influences the value a dichotomous variable will take,
controlling for the other independent variables in the model.
Verifying that dependent variables are statistically significant explanations for
variations in the dependent variable. Finally, we must look at Figure 4 one more time
to examine each independent variable's significance statistic. The odds ratio statistics
we have just analyzed are based on a fairly small sample (the first table in the SPSS
output for this regression model--not shown above--indicates that fewer than 200
students were included in the sample). We need to figure out whether or not these
statistics that show an increase or decrease in the likelihood of a given student voting
are reliable. In other words, we want to know if the odds-ratios in Figure 4 that show
our independent variables affecting the likelihood of voting are possibly due to
random chance because these statistics were generated from a sample rather than a
survey of the entire population.
Could the relationships we in Figure 4's odds ratios (as listed in the the Exp(B)
column) be due solely to chance? The significance statistics in Figure 4 show that
answer to this question is clearly yes for both the gender (Q1) and political ideology
(Q16) variables. We cannot say with a high degree of confidence (better than 95
percent certainty) that the relationships we found between these variables and a
change in the likelihood of voting in our sample would hold true in the population as a
whole because there is more than a .05 chance that our observed relationship between
voting and these two variables is due to random survey error.
For GPA (Q9), the significance statistic indicates that the probability that a full grade
increase in GPA actually corresponds to no increase (or a decrease) in the likelihood
of a student voting is essentially nonexistent. Our odds-ratio is statistically significant
because there is only one chance in a thousand that we have observed a relationship in
our sample that would not be found in the larger population from which our sample
was drawn.

Оценить