TX_1~ABS:AT/ADD:TX_2~ABS:AT


133 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

ReseaRch aRticle

The Use of Tobit and Logistic Regression Models to Study 
Factors that Affect Blood Pressure in Cardiac Patients
Bekhal S. Sedeeq, Banaz W. Y. Meran

Department of Statistics, College of Administration and Economics, Salahaddin University-Erbil, Kurdistan Region - F.R. Iraq

ABSTRACT

This research studies the factors that affect blood pressure in cardiac patients using the Tobit and logistic regression models. The 
data have been collected, from 500  patients with heart disease in hospital – heart center – Erbil. The two levels of blood pressure, 
low and high blood pressures, were taken from the patients as dependent variables plus other, independent variables (gender, age, 
urea, cholesterol, creatinine, and weight). The research shows that the median of blood pressure by means of arterial pressure (MAP) 
equation contains each of high and low blood pressures differently. This is due to the threshold value of 99.33, equal to,12/8 mmHg, 
which represent a normal level of a human blood pressure. The aim of this research is to explain the main concepts and processes 
of Tobit regression analysis (censored and truncated) and logistic regression analysis, which are used for predicting the factors of 
independent variables that have more effect on the response variables, for example, blood pressureand to compare the outcomes of the 
two models (Tobit regression and Logistic regression) in order to determine which of the models best fits our data in which AIC and 
BIC are used. The data analysis of this research shows that the logistic regression model best fits our data, as compared with the Tobit 
regression model. The data analysis has been achieved using statistical packages in R programming, MATLAB and Statistical Package 
for the Social Sciences (SPSS) V.26.

Keywords: Akaike information criterion, Bayesian information criterion, censoring, logistic regression model, Tobit regression model, truncation

INTRODUCTION

The health sector is one of the most vital sectors that undertake the task of providing health services to all members of society through health institutions to protect 
and improve society and achieve the well-being of its members. In 
fact, one of the components of building health in society is to ward 
off all diseases, and at the lead of these diseases is cardiovascular 
disease, which is one of the problems that challenge medicine.

Cardiovascular disease, the leading cause of death 
worldwide, is greatly exacerbated by high blood pressure. 
Around,54% of strokes-and 47% of coronary heart disease 
occur worldwide as a result of high blood pressure. High 
blood pressure is common medical issue that becomes more 
prevalent as people become older.[1]

There have been several statistical studies of individual 
data that include observations in which a dependent variable 
is equal to 0, for a digit of observations in the dataset. This 
behavior is known as censored or truncated data.[2]

Since James Tobin’s work, the Tobit regression has received 
a huge and diverse amount of theoretical interest (1958). Its 
use in practical applications has been developed in fields such 
as economics, biology, finance, and medicine. Tobit regression 
can be seen as a linear regression model where only the data, 
on the dependent variable are incompletely, observed.[3]

Logistic regression is a statistical, method for examining 
the relationship among a dependent variable of a nominal 
level and one or other independent variables so that those 
independent variables are from any type of measurement 
level.[4]

Logistic regression is investigation of the best method 
in the case of the binary dependent variable, according to 
Dayton[5] logistic regression, which is a categorical data 
statistical modeling method. It is the usages of the same logic 
as ordinary least squares regression.

This study, aims at:
1. Utilizing both the Tobit and the logistic regression models.

Corresponding Author: 
Banaz W. Y. Meran, Department of Statistics, College of 
Administration and Economics, Salahaddin University-Erbil, 
Kurdistan Region - F.R. Iraq.  
E-mail: banaz.yaqoob@su.edu.krd

Received: July 18, 2022 
Accepted: August 12, 2022 
Published: November 20, 2022

DOI: 10.24086/cuesj.v6n2y2022.pp133-140

Copyright © 2022 Banaz W.Y Meran, Bikhal S. Sedeeq. This is an open-access article 
distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0).

Cihan University-Erbil Scientific Journal (CUESJ)



Meran and Sedeeq: Tobit and logistic regression models

134 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

2. Knowing which factors of independent variables is more 
effecter on the dependent variable (blood pressure) using 
both models.

3. Comparing the results of the two models to determine which 
one best fits our data that are using Akaike information 
criterion (AIC) and Bayesian information criterion (BIC).

LITERATURE REVIEWS

Shirafkan et al.[6] estimated the potential of Tobit regression, 
as a method to study time to onset of cytomegalovirus in 
renal recipient transplants. The results of this study revealed 
that the age of patients was influenced by the time of onset 
of disease. Consequently, the bigger the age, the shorter time 
required to-start infection.

Karim et al.,[7] Tobit model and the traditional regression 
model were compared on data taken from patients with kidney 
disease, and through using some statistical measures, it was 
concluded that Tobit’s model is preferable than the traditional 
regression model for this type of data.

Taleb et al.[8] used a binary logistic regression model, 
estimating the coefficients of this model the least squares 
method for people with heart disease. The aim of the 
investigation was to compare the actual reasons of death with 
the estimated causes of death. The binary logistic regression 
model concluded that smoking was the leading cause of death.

Rambeli et al.[9] used the binary logistic model for a study 
the aim of which was to identify the factors that influence 
a teacher’s decision to remain in their professional life. The 
result showed that the income was considered as a key factor 
in teacher remaining committed to the profession.

METHODOLOGY

This research contains three parts: The first part of this 
research included the introduction, methodology, aim of the 
study, and review of the literature. The second part of the 
research included the basic concepts of the two models and 
the last part contains data applications, implementation of two 
models, and interpretation of the results.

Tobit Regression Model

Regression analysis is one of the most important ways to 
find out the significant effect between the dependent variable 
and explanatory, variables but sometimes, the dependent 
variable is constrained to the threshold point. In this situation, 
the use of the traditional regression model is biased. With 
this kind of data, using the Tobit regression model is the 
best option. Logistic-regression, it is used as in this, study. 
The analysis of the Tobit regression means the, statistical 
method used to examine the relationship between the limited 
dependent variable and explanatory variables of any type. The 
analysis in this situation is named a Tobit regression. A limited 
dependent variable is the one whose range of potential values 
is constrained in some significant way.[10] Limited dependent 
variable models contain: a – Censoring, in which some data 
are lost but other data are present for certain persons in a data 
collection and b – truncation, in which some individuals are 
purposefully removed from observation.[11]

Tobin used a regression model based on household 
expenditures that particularly took into account the fact 
that (his regression model’s dependent variable) cannot be 
negative. Tobin coined the phrase “model of constrained 
dependent variables” to describe his approach, a term invented 
by Goldberger (1964) for the reason that they resemble Probit 
models. These models are frequently denoted to as truncated 
or censored models. If the observations are lost outside of a 
carefully defined range and are censored, the model is called a 
truncated model.[12] The structural Tobit model

y Xi
* � �� ei�

Where, ei ~ N(0, σ
2). y* is a latent variable that is observed y, shown 

by the following equation y
y if y

if yi y
�

�
�

�
�
�

��

* *

����
*

��� ���� �����

���� ������

�
� �

When assume that τ = 0 in the standard Tobit model, data 
are censored next to a value of 0. For use there is:[13]

y
y if y

if y
i �

�
�

�
�
�

��

* *

*

��� ���� �����

���� ���� ������

0

0 0

As it has been shown, the probability function is censored

L
y µ

i

N di di

�
��

�
�

�
�
�

�

�
�

�

�
� �

��
�
�

�
�
�

�

�
�

�

�
��

�

�
1

1
1

�
�

�
�


�

�

In the conventional Tobit model, one groups τ = 0 and 
parameterize µ as (X

i
β). That is Tobit model’s likelihood 

function:

L
y X X

i

N

i i

di

i

di

�
��

�
�

�
�
�

�

�
�

�

�
� �

�
�
�

�
�
�

�

�
�

�

�
��

�
1

1
1

�
�

�
�

�
�



The Tobit model’s log likelihood function is

ln

ln

L

d
y X

d
Xi

N i
i i

i
i

�

� �
��

�
�

�
�
�

�

�
�

�

�
� �

�� � � �
�
�

��
�

1 1 1

ln ln� �
�

�

�
�

�
��
�

�

�
�

�

�
�

�

�




�




�







	




The overall log, likelihood is split into two components. 
The first component corresponds to a traditional regression 
for an uncensored observation, while the, second component 
represents the associated likelihood of censoring the test.[14]

Standard Tobit model

Censored regression method has been very widespread as 
a standard Tobit model since Tobin (1958) first presented it to 
evaluate the relationship between household expenditure and 
household income.

y i n'* ����� , , .� � � �x ui i� 1 2

y �y ������if��������y � n� �* *

y � �if�������y n� �0��������� *

Where, u
i
 is identical independent distribution (iid). 

Drawings from N(0, σ
u
).

y
i
 and X

i
 are observed in the sample, but the yi

*  is unobserved 

if � *y i2 < 0. The likelihood function for this model is:
[12]



Meran and Sedeeq: Tobit and logistic regression models

135 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

L
x y xi i i� �

�

�
�

�

�
�

�

�
�

�

�
�

��

�
�

�

�
�� �

0 1

1
1

� �
' '�
� �

�
�

Censoring

A regression, model is said to be-censored when the recorded, 
data on the dependent variable (the response) cutoff outside 
a certain-range with multiple observations at the-endpoints 
of that range. When the data are censored, variation in the 
observed dependent variable will underestimate the impact of 
the regression on the actual dependent variable. As a result, 
coefficient estimates from standard ordinary least squares 
regression employing censored data are often biased toward 
zero.[15]

Truncation

Truncated data, or missing data, are discovered when an 
observation is not reported despite of whether it is below or 
above a specific level that differs. In actuality, these are known as 
left and right truncation, respectively. The truncation effect can 
also happen when only a small portion of a larger population is 
represented in the sample data. In addition, the response variable 
in the model is truncated if observations are not possible while 
taking values inside a particular range. Consequently, when the 
dependent variable is within that range, neither the dependent 
nor the independent variables are observed.[2]

Logistic Regression Model

What distinguishes a logistic regression model from the, linear 
regression model, in logistic regression, could be the result 
variable which is either binary or, dichotomous. This difference 
among logistic and linear regression is reflected both in the 
choice of a parametric model and it assumptions, whereas the 
methods used in a logistic, regression study follow the same 
fundamental principles, as linear regression.[16,17]

Logistic regression model circuitously models the response 
variable created on probabilities linked with the digits of 
the dependent variable y. We will use P(X) to represent the 
possibility of a response when y = 1. Furthermore, we will 
define, 1–P(X) which represents the possibility of a response 
when y = 0. These probabilities are written as follows:

P X P Y X X Xk� � � � �� �1 1 2| , ,

1 0 1 2� � � � � �� �P X P Y X X Xk| , ,

The logistic regression equation and the regression 
equation with a straight line, equation 
( Y X X Xk k� � � � ��� � � �1 1 2 2 . ) are related by the formula 

below. The logistic regression equation form is rewritten as 
such:

Logit P X
P X

P X
X X Xe k k� log .� � �

� �
� � �� �

�

�
�
�

�

�
�
�
� � � � ��

1
1 1 2 2�� � � �

The logistic regression model can be calculated using the 
formula below.[17,18]

P X
X X kXk

X X kXk
( )

.
�

�

� � ���� �

� � �� �� �
e
e

��

��

� � �

� � �

1 1 2 2

1 1 2 21

Estimate the Parameters of Logistic 
Regression Model

The approach of maximum likelihood estimation will be used. 
The log likelihood is given as:[19-21]

( )yi 1 yi
1

(  ,   ) [1 ( )]
n

i

L P X P X −

=

α β = −∏

We’ll use the log likelihood method for estimation:

logL logP X P X
i

( �, ) ( ) ( ( )� � � � �
�

�
��

�

�
��

�
�

�
�
�

�

� �
� �

1 1

1
n

i

i

n

iy n y log
��
�
�

�



� � �

�

�
��

�

�
��

�
� �
� �U P L P X

i

( )
( , )

/ ( ) /( (
� ���
P

y n y P X
i

n

i

n

i

1 1

1 )))
�

�
�
�

�

�
�
�

Cox and Snell R2 statistic

In the logistic regression model, the determination coefficient 
R2 used to determine the fittingness of the proposed regression, 

Table 1: Five hundred patients sample

ID Y: BP X1: Gender X2: Age X3: Urea X4: Cholesterol X5: Creatinine X6: Weight

1 88 (0) 1 70 68 127 1.26 89

2 76 (0) 2 40 41 117 0.97 96

3 99.33 1 44 34 201 0.59 74

4 85 (0) 2 43 33 221 0.67 107

5 117 1 55 54 226 1.12 88

6 92 (0) 1 67 31 173 0.72 76

7 96.67 1 47 43 165 0.78 65

497 107.67 1 54 31 160 0.87 85

498 93.67 2 63 16 109 0.6 69

499 106.67 1 58 25 100 0.73 67

500 118.33 1 69 38 134 1.11 65

BP: Blood pressure



Meran and Sedeeq: Tobit and logistic regression models

136 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

models for the study data is changed by the R2 Nagelkerke, 
Cox, and Snell R2 computation statistics.

Can be calculated as: �
( / )

R Cox�and�Snell2 0
1

2

1� �
�

�
�

�

�
�

L
L

n

L
0
: Maximum likelihood for constant in the model.

L
1
: Maximum likelihood independent variables in the model

n: Sample size[22]

And, can be calculated as:

R
R

n
2

2

0
21

Nagelkerke
cox snell

L
�

�
� �&�

( / )

The Hosmer-Lemeshow test

The Hosmer and Lemeshow test is a widely-used test for 
determining a model’s quality of fit. It accepts any number of 
explanatory-variables, which can be continuous or categorical. 
It is used to test the hypothesis:

H
0
: The model is adequate for data

H
1
: The model is not adequate for data

The model will be a useful model if (Hosmer and 
Lemeshow) static is >0.5.[22]

HL
o E

Nk
i

n
i i

i i i

�
� ��

�

�
�� �
�� ��2

2

1

2

1

Wald statistic

The Wald test is used to determine whether or not the effect 
of the logistic regression coefficient on the independent 
variables.[23]

The Wald statistic is calculated according to the following formula: 
2

2       
.

W
S Eβ

 β
=  
  

The classification table

The use of classification tables is one of the ways to check 
the quality of matching the model with the data. This method 
depends on creating tables that put the number of cases that 
have the desired trait or the cases that do not have the desired 
trait and that were categorized correctly or incorrectly. The 
concept behind using the analysis is it leading to better results 
if the model is data compatible.[24]

The AIC

AIC as a model selection criterion for assessing actual data, 
it has played a key role in solving issues in a wide range of 
fields, and the model by the lowermost AIC is chosen as the 
best model.[25]

AIC L K�� �� log *2 2

The BIC

In statistical model selection, the BIC is unique of the best 
well-known and commonly used tools. BIC is calculated for 
each of the models, and the model with the lower BIC value is 
selected as the best model.[26]

 2 log 2 * log *BIC L N K=− +

DATA ANALYSIS

In this study, the two models of Tobit and logistic regression 
models have been applied on a sample taken from 500 patients 
with heart disease and two levels of blood pressure, high and low 
blood pressure, in hospital – heart center – Erbil. Blood pressure 
is taken from the patients as response variables and some 
independent-variables (gender, urea, age, cholesterol, creatinine, 
and weight). The study found that the average of blood pressure 
by means of arterial pressure (MAP) equation contains each high 
and low blood pressure differently because the threshold point 
was determined to be 99.33. To take the best model for our data 
in the study, two statistical measures (AIC and BIC) were used.

Note: All assumptions and tests related to Tobit and 
logistic regression models have been applied before we started 
the data analyses in this study.

Data Description for Tobit Regression 
Analysis

In this study, the data are gathered from 500 patients with 
heart diseases, and the two levels of blood pressure; high and 
low blood pressure were taken from patients as dependent 
variables and the variables: Gender, age urea, cholesterol, 
creatinine, and weight as the independent variables (Table 1). 
The researcher set that the medium of blood pressure by MAP 
equation contains each highest and lowest blood pressure 
differently because the threshold point was determined to be 
93.33,

In regards, the dependent variable in this study has been 
defined as:

Y=Y* Y*>93.33

Appropriately, the limitation threshold point Y = 93.33 
was regarded the limitation threshold point y = 0, in 
accordance with the stated theoretical presentation, and the 
model will be referred to as the following model:

Y=Y* Y*>93.33

Y=0  Y*<=93.33

The explanatory variables in this study are the follows:
X

1
: Gender (male and female)

X
2
: Age measured by year

X
3
: Urea (mg/dL)

X
4
: Cholesterol (mg/dL)

X
5
: Creatinine (mg/dL)

X
6
: Weight measured by kg

The explanation of the variables is presented in Table 2 in 
which contains independent variables and dependent variable 
and the table views the maximum, minimum, mean, and 
standard deviation of data.

Application of Tobit regression model analysis (Censored and 
Truncated)

Because of the researcher initially checked all the necessary 
assumptions that must be present in the data before starting to 
analyze the data, and also set a unified standard for the data 
in this study, it becomes clear for us that there is no problem 
in terms of our data and we can use the data for analyzing.



Meran and Sedeeq: Tobit and logistic regression models

137 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

Censored regression model

First, let’s start with censored-regression model are: Censored 
(formula = Y~ X, -left = 0, right = Inf., data= data 1) Total 
(n = 500 observation, left censored = 132 observation, 
Uncensored = 368 observation, left censored (Y < 99.3 then 
Y* = 0: Observation).

Table 3 presents the results of the exact regression 
model. The coefficients of the independent variables sex, 
age, cholesterol, creatinine, and weight are positive because 
the variables have a positive relationship with the dependent 
variable (blood pressure), while the coefficient of the 
independent variable urea is negative because the variable has 
a negative relationship with the dependent variable (blood 
pressure). According to the findings in Table 3, the variables 
age, cholesterol, and creatinine all significantly affect blood 
pressure. The summarized fit to the censored regression model 
is log-likelihood = −2125,549, AIC = 4265.1, BIC = 4294.6. 
The score for the best model is determined by the lowest value 
for AIC and BIC.

Truncated regression model

In this case, the number of observations turns to 368 due to a 
truncation.

Table 4 shows the results from truncated regression 
model. Those coefficients of the independent variables such as 
gender, age, cholesterol, urea, and weight are positive for the 
reason that the variables take a positive relationship with the 
dependent variable (blood pressure) whereas the coefficient 
of the independent variable creatinine is negative because 
has a negative connection with the dependent variable (blood 
pressure). It is through the conclusion in Table 4 that only the 
urea variable has the effect on blood pressure.

Logistic Regression Analyses

In this part, we use binary logistic regression since the dependent 
variable in this study is blood pressure and the researcher has 
taken the average of blood pressure by MAP equation which 
contains each of the high and low blood pressure differently 
because of the threshold points which was determined to 
be 93.33. The patient whose blood pressure is >93.33 is 
considered to be infected and takes the worth of one while the 
patient whose blood pressure is ≤93.33 is considered to be 
uninfected and takes the value of 0, and the other variables are 
gender, age, cholesterol, urea, creatinine, and weight.

Y: The dependent variable has a binary response code: 

Y
Infected
Uninfected

�
�
�
�

1

0

The independent variables are the same as the variables 
written above.

Application of binary logistic regression analysis

Let’s start with the outcome of the classification table starting 
with the zero stage in which the model is free of independent 
variables (only the constant).

Table 5 represents the baseline model, which is a model 
without our explanatory variables. The overall right percentage 
was 73.6% which refers the model’s overall explanatory 
strength . The initial log likelihood function (-2 log likelihood 
function)= 577.2.

Omnibus test of logistic model coefficients

Based on the model coefficients in omnibus tests, we find that the 
Chi-square tests are to illustrate if there is an important variance 
between the factors of the nil model and the current model.

Table 2: Descriptive statistics of variable

Variables Minimum Maximum Mean SD

BP 70.33 141.00 101.23 10.97

Gender 1 2 1.33 0.47

Age 17 86 57.73 11.24

Urea 11 198 40.11 20.53

Cholesterol 63 326 167.36 45.02

Creatinine 0.11 10.70 1.58 1.195

Weight 48 135 78.83 13.47

BP: Blood pressure, SD: Standard deviation

Table 3: Censored regression model

Coefficients Estimate SE t Pr (>t)

Constant −127.94562 25.77314 −4.964 6.89e-07*** 

Gender 3.57944 5.59007 0.640 0.522

Age 1.91136 0.24742 7.725 1.12e-14***

Cholesterol 0.25688 0.05688 4.516 6.30e-06***

Urea −0.11037 0.12724 −0.867 0.386

Creatinine 12.21152 2.09200 5.837 5.31e-09***

Weight 0.31290 0.19800 1.580 0.114

*** mean P ≤ 0.001

Table 4: Truncated regression model

Coefficients Estimate SE t Pr (>t)

Constant 98.7995674 4.7143913 20.9570 <2e-16*** 

Gender 0.6703624 0.9215182 0.7275 0.46695

Age 0.0645897 0.0448466 1.4402 0.14980

Cholesterol 0.0015114 0.0094141 0.1605 0.87245

Urea 0.0520044 0.0202811 2.5642 0.01034*

Creatinine −0.2198352 0.3154896 −0.6968 0.48592

Weight 0.0040456 0.0340886 0.1187 0.90553

SE: Standard error; * mean P ≤ 0.05 and  *** mean P ≤ 0.001

Table 5: Classification table shows that the model has a constant 
bound zero step

Observed Predicted

Y Percentage correct

Uninfected Infected

Step 0

Y

Uninfected 0 132 0.0

Infected 0 368 100.0

Overall percentage 73.6



Meran and Sedeeq: Tobit and logistic regression models

138 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

Table 6 displays that the current model is meaningfully more 
suitable than the null model. Model coefficients omnibus test offers 
significant reduction in the −2 log likelihood value = 426.572= 
as compared to −2 log likelihood value =577.2 of the null 
model. This indicates that the Chi-square values for step, block, 
and model are all the same. P < 0.05 illustrating that the model’s 
accuracy increases when explanatory variables are included. 
Furthermore, the Chi-square is significant (χ2 = 150.628, df = 6, 
P < 0.05). As a result, our new model is bested.

Hosmer and Lemeshow test

Being a goodness of fit test for logistic regression, it fits the 
data model.

Table 7 explains since P = 0.226, it is larger than the level 
of significance at 5%. We may well conclude that the data are 
will be suitable to the model.

Cox and Snell R2, Nagelkerke’s R2

The Cox and Snell R2, Nagelkerke’s R2 values are utilized to 
estimate the model’s fit to the data.

As the Nagelkerke’s R2, Cox and Snell R2 values given 
in Table 8 are examined, the ratios of interpretation of 
the independent variables over the dependent variable are 
shown. The value of Nagelkerke R2 is the modified form for 
the Cox and Snell R2 coefficient. Giving to the outcomes 
shown in Table 8, it is seen that the dependent variables 
determine 26% of the variance in the independent variables 
according to the value of Cox and Snell R2, and 38% 
according to the value of Nagelkerke R2. The amount of the 
−2 log likelihood statistic is 426.572 in model summary for 
the whole model.

Table 9 is corresponding to Table 5 but it is based on 
the model that contains our explanatory variables. The total 
percentage of correct was 79.2% which replicates the model’s 
overall explanatory strength. Classification table contains the 
constant term and the rest of the predictors, that is, 91.6% 
were correct for the infected of blood pressure and correctly 
classified. This table shows how many events were correctly 
predicted (59 cases were observed to be uninfected and were 
correctly predicted to be uninfected; 337 cases were observed 

to be infected and were correctly predicted to be infected) and 
how many were not (73 cases are observed to be uninfected 
but are predicted to be infected; 31 cases are observed to be 
infected but are predicted to be uninfected).

Variables in the equation logistic-regression model

The variables in the equation logistic regression are the 
most essential of all the outputs. This table must be studied 
carefully since it contains the answers to our questions about 
the common relationship between all variables.

Table 10 shows the values of Wald test that represents the 
parameter of the test value of the model and it appears that 
the variables (age, urea cholesterol, and creatinine) represent 
the significant variables in the research. It is by associating the 
P-value with the level of significant (0.05) that the p-value 
represents the significance of the effect of the variable on 
the patient condition, that is, it is significant when P < 0.05 
was considered for the variable under test. It shows that the 
variables (gender and weight) are not significant variables 
in the study, and it is through comparing the P-value with 
the level of significant (0.05) that the P-value represents the 
non-significance of the effect of the variables on the patient 
condition, that is, it is no significant when P > 0.05 was 
considered.

As far as Table 10 is concerned, the logistic computation 
coefficients that may be utilized to build a predictive equation 
could be:

Y �
� �

� � �
�

�
��

�

�
�� � � � ��log

p x

p x
x x xk k1 1 1 2 2

� � � �

The effect factors for blood pressure in cardiac patients 
can be ranked as follows based on the value of the odds ratio. 
Likewise, we can write the logistic regression computation 
with just significant variables:

Logistic equation (Model): Y = −7.689 + 0.080 
Age - 0.016 Urea + 0.013 Cholesterol + 1.451 Creatinine

Table 10 shows, Exp(β) = eb represents the ratio changed 
in the odds of the event of importance for a one unit variation 
in the predictor.

The value of Exp(β) for the variable gender indicates 
that when the gender changes from the value 0 (female) to 
the value 1 (male), the probability of disease blood pressure 
in patients with heart disease increases because the value of 
Exp(β) is >1. Such incomes indicate that the blood pressure of 
males is higher than females.

Table 6: Omnibus test

Step 1 χ2 df Significant

Step 150.628 6 0.000

Block 150.628 6 0.000

Model 150.628 6 0.000

Table 7: Hosmer and Lemeshow test

Step χ2 df Significant

1 10.595 8 0.226

Table 8: Summary of logistic regression

Step −2 Log 
likelihood

Cox and Snell R2 Nagelkerke R2

1 426.572a 0.260 0.380

Table 9: Classification tables

Observed Predicted

Y Percentage correct

Uninfected Infected

Step 1

Y

Uninfected 59 73 44.7

Infected 31 337 91.6

Overall percentage 79.2



Meran and Sedeeq: Tobit and logistic regression models

139 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

The odds ratio for the variable age is >1, Exp(β) = 1.083. 
This means that each additional rise of 1 year in age is in touch 
with the increase in the odds infected of blood pressure in 
cardiac patients.

The odds ratio for the variable urea is <1. This means that 
each additional increase of one unit in urea is associated with 
decrease in the odds infection of blood pressure in cardiac 
patients with 0.984 times.

The odds ratio for the variable cholesterol is >1. This 
means that each additional increase of one unit in cholesterol 
is related to the increase in the odds of blood pressure in 
cardiac patients with 1.013 times.

The odds ratio for the variable creatinine is >1. This 
means that each additional increase of one unit in creatinine is 
related to the increase in the odds infection of blood pressure 
in cardiac patients with 4.267 times.

The odds ratio for the variable weight is >1. This 
means that each additional increase of one unit in weight is 
associated-with the increase in the odds infection of blood 
pressure in cardiac patients with 1.011 times.

DISCUSSION

There are different techniques for comparing the analysis 
of two or more models; however, the AIC and BIC criteria are 
two that may be worth considering.

Table 11 shows a comparison between three regression 
models censored, truncated, and logistic, for choosing the most 
fit model to our data of blood pressure in cardiac patients, 
the AIC and BIC values with the least values are chosen. The 
results display that the logistic regression model is better and 
more suitable rather than truncated regression and censored 
regression for our data, because it’s AIC equal to 591.2 and 
BIC = 620.7 are the lowest values contrast with Tobit models 
(censored and truncated).

CONCLUSION

It has been concluded the following:
1. In the censored, regression model, the explanatory, 

variables (age, cholesterol, and urea) significantly 
impacted on blood pressure.

2. The results, from truncated regression model show, that 
only the urea variable has the effect on blood pressure.

3. According, to the results classification table, logistic 
model is, correctly classifying the consequences for 79.2% 
of the cases, compared to 73.6% in the null, model.

4. According to the Hosmer-Lemeshow test, our data fit the 
logistic regression based on a χ2 = 10.595 and P-value 
greater, than a significant level.

5. Wald’s test showed that the variables of age, creatinine, 
cholesterol, and urea, respectively, contributed 
significantly to the prediction, depending on the P-value 
(0.000 < 0.005). The variables that do not have a 
significant, effect are weight and gender.

6. It was concluded, that the logistic regression model for 
the sample under study or for our data is better, than the 
censored regression model and the truncated regression 
model after comparing, their AIC and BIC values.

REFERENCES
1. C. Y. Wu, H. Y. Hu, Y. J. Chou, N. Huang, Y. Chou and C. Li. 

High blood pressure and all-cause and cardiovascular 
disease mortalities in community-dwelling older adults. 
Medicine(Baltimore), vol. 94, no. 47, p. e2160, 2015.

2. W. H. Greene. Censored Data and Truncated Distributions, SSRN 
Electron. J, 2005.

3. M. H. Odah, B. K. Mohammed and A. S. M. Bager. Tobit 
regression model to determine the dividend yield in Iraq. LUMEN 
Proceedings, vol. 3, pp. 347-354, 2018.

4. J. S. Cramer. The Origins of Logistic Regression: Tinbergen Institute 
Discussion Papers, 2002.

5. C. M. Dayton. Logistic regression analysis. Stat, vol. 474, 
p. 574, 1992.

6. H. Shirafkan, J. Yazdani-Charati, S. A. Mozaffarpur, S. Khafri, 
R. Akbari and A. A. Pasha. Application of tobit model in time 
until Cytomegalovirus infection in kidney transplant recipients. 
Acta Medica, vol. 32, p. 1237, 2016.

7. R. M. H. Karim and S. M. Salh. Using tobit model for studying 
factors affecting blood pressure in patients with renal failure. 
UHD Journal of Science and Technology, vol. 4, no. 2, pp. 1-9, 
2020.

8. H. R. Talib and S. A. Mazloum. The use of binary logistic 

Table 10: Variables in the equation

Step 1 β SE Wald df Significant Exp(β) 95% CI for Exp(β)

Lower Upper

Gender 0.108 0.264 0.168 1 0.682 1.114 0.664 1.868

Age 0.080 0.012 41.674 1 0.000*** 1.083 1.057 1.110

Urea −0.016 0.007 4.692 1 0.030* 0.984 0.970 0.998

Cholesterol 0.013 0.003 21.265 1 0.000*** 1.013 1.007 1.019

Creatinine 1.451 0.276 27.595 1 0.000*** 4.267 2.483 7.333

Weight 0.011 0.009 1.327 1 0.249 1.011 0.992 1.030

Constant −7.689 1.294 35.304 1 0.000 0.000

CI: Confidence interval, SE: Standard error, * mean P ≤ 0.05 and *** mean P ≤ 0.001

Table 11: Comparison

Tobit regression 
model (censored)

Truncated 
regression model

Logistic 
regression model

AIC=4265.1 AIC=2574.6 AIC=591.2

BIC=4294.6 BIC=2602.0 BIC=620.7

AIC: Akaike information criterion, BIC: Bayesian information criterion



Meran and Sedeeq: Tobit and logistic regression models

140 http://journals.cihanuniversity.edu.iq/index.php/cuesj CUESJ 2022, 6 (2): 133-140

regression method to analyze the factors affecting heart disease 
deaths: An applied study on a sample of patients in Dhi Qar 
Governorate. Journal of Al-Rafidain University College College for 
Sciences, 2020, no. 46, 2020.

9. N. Rambeli, E. Hashim, F. C. Leh, N. S. Hudin, M. F. Ramli, M. C. 
Mustafa, et al. Decision to leave or remain in the career as early 
childhood educator: A binary logistic regression model. Review 
of International Geographical Education Online, vol. 11, no. 5, 
pp. 450-456, 2021.

10. G. S. Maddala. Limited-Dependent and Qualitative Variables in 
Econometrics. New York, NY: Cambridge University, 1983.

11. N. M. Ahmed. Limited dependent variable modelling (truncated and 
censored regression models) with application. The Scientific Journal 
of Cihan University Sulaimanyia, vol. 2, no. 2, pp. 82-96, 2018.

12. T. Amemiya. Tobit models: A survey. Journal of Econometrics, 
vol. 24, no. (1-2), pp. 3-61, 1984.

13. J. S. Long, Regression Models for Categorical and Limited 
Dependent Variables. Thousand Oaks, CA: Saga Publication, 1997.

14. A. Flaih, J. Guardiola, H. Elsalloukh and C. Akmyradov. Statistical 
inference on the ESEP tobit regression model. Journal of Statistics 
Applications and Probability Letters, vol. 6, no. 1, pp. 1-9, 2019.

15. K. Y. Chay and J. L. Powell. Semiparametric censored regression 
models. Journal of Economic Perspectives, vol. 15, no. 4, pp. 29-42, 
2001.

16. D. Hosmer Jr., S. Lemeshow and R. Sturdivant. Applied Logistic 
Regression. vol. 398. New York: John Wiley and Sons, 2013.

17. A. M. Khudhur and D. H. Kadir. An application of logistic 
regression modeling to predict risk factors for bypass graft 

diagnosis in Erbil. Cihan University-Erbil Scientific Journal, vol. 6, 
no. 1, pp. 57-63, 2022.

18. N. M. M. Abd Elsalam. Binary logistic regression to identify the 
risk factors of eye glaucoma. International Journal of Sciences 
Basic and Applied Research, vol. 23, no. 1, pp. 366-376, 2015.

19. N. S. K. Barznj. Using logistic regression analysis and linear 
discriminant analysis to identify the risk factors of diabetes. 
Zanco Journal of Humanity Sciences, vol. 22, no. 6, pp. 248-268, 
2018.

20. S. Menard. Applied Logistic Regression Analysis. vol. 106, 
New York: Sage, 2002.

21. N. H. Mahmood, R. O. Yahya and S. J. Aziz. Apply binary logistic 
regression model to recognize the risk factors of diabetes through 
measuring glycated hemoglobin levels. Cihan University-Erbil 
Scientific Journal, vol. 6, no. 1, pp. 7-11, 2022.

22. V. Bewick, L. Cheek and J. Ball. Statistics review 14: Logistic 
regression. Critical Care, vol. 9, no. 1, pp.112-118, 2005.

23. D. Hosmer and S. Lemeshow. Applied Logistic Regression. 
New York: Johnson Wiley and Sons, 2000.

24. I. R. Soderstrom and D. W. Leitner. The effects of base rate, selection 
ratio, sample size, and reliability of predictors on predictive 
efficiency indices associated with logistic regression models, 1997.

25. S. Konishi and G. Kitagawa. Information Criteria and Statistical 
Modeling. Berlin: Springer Science and Business Media, 2008.

26. K. I. Mawlood. Using logistic regression and cox regression 
models to studying the most prognostic factors for leukemia 
patients. Qalaai Zanist Scientific Journal, vol. 4, no. 3, pp. 705-
724, 2019.