Version info: Code for this page was tested in SAS 9.3.
Interval regression is used to model outcomes that have interval censoring. In other words, you know the ordered category into which each observation falls, but you do not know the exact value of the observation. Interval regression is a generalization of censored regression.
Please note: The purpose of this page is to show how to use various data analysis commands. It does not cover all aspects of the research process which researchers are expected to do. In particular, it does not cover data cleaning and checking, verification of assumptions, model diagnostics or potential follow-up analyses.
Examples of interval regression
Example 1. We wish to model annual income using years of education and marital status. However, we do not have access to the precise values for income. Rather, we only have data on the income ranges: <$15,000, $15,000-$25,000, $25,000-$50,000, $50,000-$75,000, $75,000-$100,000, and >$100,000. Note that the extreme values of the categories on either end of the range are either left-censored or right-censored. The other categories are interval censored, that is, each interval is both left- and right-censored. Analyses of this type require a generalization of censored regression known as interval regression. Example 2. We wish to predict GPA from teacher ratings of effort and from reading and writing test scores. The measure of GPA is a self-report response to the following item:
Select the category that best represents your overall GPA. less than 2.0 2.0 to 2.5 2.5 to 3.0 3.0 to 3.4 3.4 to 3.8 3.8 to 3.9 4.0 or greater
Again, we have a situation with both interval censoring and left- and right-censoring. We do not know the exact value of GPA for each student; we only know the interval in which their GPA falls.
Example 3. We wish to predict GPA from teacher ratings of effort, writing test scores and the type of program in which the student was enrolled (vocational, general or academic). The measure of GPA is a self-report response to the following item:
Select the category that best represents your overall GPA. 0.0 to 2.0 2.0 to 2.5 2.5 to 3.0 3.0 to 3.4 3.4 to 3.8 3.8 to 4.0
This is a slight variation of Example 2. In this example, there is only interval censoring.
Description of the data
Let’s pursue Example 3 from above.
We have a hypothetical data file, intregex, with 30 observations. The GPA score is represented by two values, the lower interval score (lgpa) and the upper interval score (ugpa). The writing test scores, the teacher rating and the type of program (a nominal variable which has three levels) are write, rating and type, respectively.
Let’s look at the data. It is always a good idea to start with descriptive statistics.
proc print data = mylib.intreg_data;
var lgpa ugpa;
run;
Obs lgpa ugpa
1 2.50000 3.00000
2 3.40000 3.80000
3 2.50000 3.00000
4 0.00000 2.00000
5 3.00000 3.40000
6 3.40000 3.80000
7 3.80000 4.00000
8 2.00000 2.50000
9 3.00000 3.40000
10 3.40000 3.80000
11 2.00000 2.50000
12 2.00000 2.50000
13 2.00000 2.50000
14 2.50000 3.00000
15 2.50000 3.00000
16 2.50000 3.00000
17 3.40000 3.80000
18 2.50000 3.00000
19 2.00000 2.50000
20 3.00000 3.40000
21 3.40000 3.80000
22 3.80000 4.00000
23 2.00000 2.50000
24 3.00000 3.40000
25 3.40000 3.80000
26 2.00000 2.50000
27 2.00000 2.50000
28 2.00000 2.50000
29 2.50000 3.00000
30 2.50000 3.00000
Note that there are two GPA responses for each observation, lgpa for the lower end of the interval and ugpa for the upper end.
proc means data = mylib.intreg_data;
var lgpa ugpa write rating;
run;
The MEANS Procedure
Variable N Mean Std Dev Minimum Maximum
------------------------------------------------------------------------------
lgpa 30 2.6000000 0.7754865 0 3.8000000
ugpa 30 3.0966666 0.5708332 2.0000000 4.0000000
write 30 113.8333333 49.9427834 50.0000000 205.0000000
rating 30 57.5333333 8.3034406 48.0000000 72.0000000
------------------------------------------------------------------------------
proc sort data = mylib.intreg_data;
by type;
run;
proc means data = mylib.intreg_data;
by type;
var lgpa ugpa;
run;
--------------------------------------------- type=1 ---------------------------------------------
The MEANS Procedure
Variable N Mean Std Dev Minimum Maximum
------------------------------------------------------------------------------
lgpa 8 1.7500000 0.7071068 0 2.0000000
ugpa 8 2.4375000 0.1767767 2.0000000 2.5000000
------------------------------------------------------------------------------
--------------------------------------------- type=2 ---------------------------------------------
Variable N Mean Std Dev Minimum Maximum
------------------------------------------------------------------------------
lgpa 10 2.7800000 0.3852849 2.5000000 3.4000001
ugpa 10 3.2400000 0.3373096 3.0000000 3.8000000
------------------------------------------------------------------------------
--------------------------------------------- type=3 ---------------------------------------------
Variable N Mean Std Dev Minimum Maximum
------------------------------------------------------------------------------
lgpa 12 3.0166667 0.6336522 2.0000000 3.8000000
ugpa 12 3.4166666 0.5474458 2.5000000 4.0000000
------------------------------------------------------------------------------
Graphing these data can be rather tricky. So just to get an idea of what the distribution of GPA is, we will do separate histograms for lgpa and ugpa. We will also correlate the variables in the dataset.
proc sgplot data = mylib.intreg_data; histogram lgpa / scale = count showbins; density lgpa; run;proc sgplot data = mylib.intreg_data; histogram ugpa / scale = count showbins; density ugpa; run;
proc corr data = mylib.intreg_data; var lgpa ugpa write rating; run; The CORR Procedure 4 Variables: lgpa ugpa write rating Simple Statistics Variable N Mean Std Dev Sum Minimum Maximum lgpa 30 2.60000 0.77549 78.00000 0 3.80000 ugpa 30 3.09667 0.57083 92.90000 2.00000 4.00000 write 30 113.83333 49.94278 3415 50.00000 205.00000 rating 30 57.53333 8.30344 1726 48.00000 72.00000 Pearson Correlation Coefficients, N = 30 Prob > |r| under H0: Rho=0 lgpa ugpa write rating lgpa 1.00000 0.94878 0.62057 0.53551
Analysis methods you might consider
Below is a list of some analysis methods you may have encountered. Some of the methods listed are quite reasonable, while others have either fallen out of favor or have limitations.
- Interval regression – This method is appropriate when you know into what interval each observation of the outcome variable falls, but you do not know the exact value of the observation.
- Ordered probit – It is possible to conceptualize this model as an ordered logistic regression with six ordered categories: 0 (0.0-2.0), 1 (2.0-2.5), 2 (2.5-3.0), 3 (3.0-3.4), 4 (3.4-3.8), and 5 (3.8-4.0).
- Ordinal logistic regression – The results would be very similar in terms of which predictors are significant; however, the predicted values would be in terms of probabilities of membership in each of the categories. It would be necessary that the data meet the proportional odds assumption which, in fact, these data do not meet when converted into ordinal categories.
- OLS regression – You could analyze these data using OLS regression on the midpoints of the intervals. However, that analysis would not reflect our uncertainty concerning the nature of the exact values within each interval, nor would it deal adequately with the left- and right-censoring issues in the tails.
Interval regression analysis
We will use proc lifereg to run the interval regression analysis. We list the variable type on the class statement. We enclose both lgpa and ugpa in parentheses on the model statement before the equals sign to indicate that these variables are the outcome variables. We list write, rating and type as the predictor variables. We use the d=normal option to specify the distribution as normal.
proc lifereg data = intreg_data;
class type;
model (lgpa ugpa) = write rating type / d=normal;
run;
The LIFEREG Procedure
Model Information
Data Set MYLIB.INTREG_DATA intreg_data dataset
written by Stat/Transfer
Ver. 10.1.1655.0406
Dependent Variable lgpa
Dependent Variable ugpa
Number of Observations 30
Noncensored Values 0
Right Censored Values 0
Left Censored Values 0
Interval Censored Values 30
Number of Parameters 6
Name of Distribution Normal
Log Likelihood -33.12890521
Number of Observations Read 30
Number of Observations Used 30
Class Level Information
Name Levels Values
type 3 1 2 3
Fit Statistics
-2 Log Likelihood 66.258
AIC (smaller is better) 78.258
AICC (smaller is better) 81.910
BIC (smaller is better) 86.665
Algorithm converged.
Type III Analysis of Effects
Wald
Effect DF Chi-Square Pr > ChiSq
write 1 9.7541 0.0018
rating 1 2.1314 0.1443
type 2 18.7076 ChiSq
Intercept 1 1.8136 0.5011 0.8315 2.7957 13.10 0.0003
write 1 0.0053 0.0017 0.0020 0.0086 9.75 0.0018
rating 1 0.0133 0.0091 -0.0046 0.0312 2.13 0.1443
type 1 1 -0.7097 0.1668 -1.0367 -0.3827 18.10
- The model information is given first. Among other things, it includes a listing of the outcome variables, the number of cases in the dataset, the type of distribution used, the log likelihood, and the number of cases read and used in the analysis.
- Information on the variable listed on the class statement is listed next. This indicates that the variable type has three levels.
- The Model Fit Summary table gives information about the model, including the -2 log likelihood, the AIC and BIC values. These values can be used to compare models.
- In the table called Type III Analysis of Effects, we see the degrees of freedom, Wald chi-square and p-values for the variables in the equation. The variable write is statistically significant, the variable rating is not, and the two degree of freedom test of the variable type is statistically significant at the .05 level.
- The ancillary statistic Scale is equivalent to the standard error of estimate in OLS regression. The value of 0.29 can be compared to the standard deviations for lgpa and ugpa, which are 0.78 and 0.57, respectively. This shows a substantial reduction. The output also contains an estimate of the standard error of sigma, as well as a 95% confidence interval.
The lifereg procedure does not compute an R2 or pseudo-R2. You can compute a rough-and-ready measure of fit by calculating the R2 between the predicted and observed values.
proc lifereg data = mylib.intreg_data; class type; model (lgpa ugpa) = write rating type / d=normal; output out = mylib.t xbeta=xb; run; ods output PearsonCorr=mylib.int_corr; proc corr data = mylib.t nosimple; var xb lgpa ugpa; run; data _null_; set mylib.int_corr; file print; if variable = "lgpa" then do; a = round((xb)**2, .0001); put "The squared multiple correlation between lgpa and the predicted value is " a; end; if variable = "ugpa" then do; b = round((xb)**2, .0001); put "The squared multiple correlation between ugpa and the predicted value is " b; end; run; The squared multiple correlation between lgpa and the predicted value is 0.6314 The squared multiple correlation between ugpa and the predicted value is 0.7107
Things to consider
See also
- SAS online manual
References
- Long, J. S. and Freese, J. 2006. Regression Models for Categorical and Limited Dependent Variables Using Stata, Second Edition. College Station, TX: Stata Press.
- Long, J. S. 1997. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks, CA: Sage Publications.
- Stewart, M. B. 1983. On least squares estimation when the dependent variable is grouped. Review of Economic Studies 50: 737-753.
- Tobin, J. 1958. Estimation of relationships for limited dependent variables. Econometrica 26: 24-36.