Simple Explanation of Statsmodel Linear Regression Model Summary

9/17/23, 11:47 AM Simple Explanation of Statsmodel Linear Regression Model Summary | by Md Sohel Mahmood | Towards Data Science
Open in app
Member-only story
Simple Explanation of Statsmodel Linear

Regression Model Summary
Statsmodel library model summary explanation
Md Sohel Mahmood · Follow

Published in Towards Data Science
7 min read · Apr 22, 2022
Listen Share More
Image by Author
Introduction
https://medium.com/towards-data-science/simple-explanation-of-statsmodel-linear-regression-model-summary-35961919868b 1/19
Regression analysis is the bread and butter for many statisticians and data
scientists. We perform simple and multiple linear regression for the purpose of
prediction and always want to obtain a robust model free from any bias. In this
article, I am going to discuss the summary output of python’s statsmodel library
using a simple example and explain a little bit how the values reflect the model
performance.
Typical model summary
For the purposae of demonstration, I will use kaggle’s Salary dataset (Apache 2.0
open source license). This dataset has two columns: years of experience and salary.
I have two two more column: Projects and People_managing.
Sample data
When we use statsmodel to use all the three variables to predict Salary, we get the
following summary result.
I am going to explain all these parameters in the summary below.
Dep variable
“Salary” which is the only dependent variable in the data.
Model and Method
OLS which stands for Ordinary Least Square. The model tries to find out a linear
expression for the dataset which minimizes the sum of residual squares.
DF residuals and DF model
We have total 30 observation and 4 features. Out of 4 features, 3 features are

independent. DF Model is therefore 3. DF residual is calculated from total
observation-DF model-1 which is 30–3–1 = 26 in our case.
Covariance type
Covariance type is typically nonrobust which means there is no elimination of data

to calculate the covariance between features. Covariance shows how two variables
move with respect to each other. If this value is greater than 0, both move in same
direction and if this is less than 0, the variables mode in opposite direction.
Covariance is difference from correlation. Covariance does not provide the strength
of the relationship, only the direction of movement whereas, correlation value is
normalized and ranges between -1 to +1 and correlation provides the strength of
relationship. If we want to obtain robust covariance, we can declare
cov_type=HC0/HC1/HC2/HC3. However, the statsmodel documentation is not that
rich to explain all these. HC stands for heteroscedasticity consistent and HC0
implements the simplest version among all.
R-squared
R-squared value is the coefficient of determination which indicates the percentage

of the variability if the data explained by the selected independent variables.
Adj. R-squared
As we add more and more independent variables to our model, the R-squared
values increases but in reality, those variables do not necessarily make any
contribution towards explaining the dependent variable. Therefore addition of each
unnecessary variables needs some sort of penalty. The original R-squared values is
adjusted when there are multiple variables incorporated. In essence, we should
always look for adjusted R-squared value while performing multiple linear
regression. For a single independent variable, both R-squared and adjusted R-
squared value are same.
Before moving to F-statistics, we need to understand the t-statistics first. T-statistics

are provided in the table shown below.
coef and std err
The coef column represents the coefficients for each independent variable along
with intercept value. Std err is the standard deviation of the corresponding
variable’s coefficient across all the data points. When using only one predicting
variable, the standard error can be obtained from this two dimensional space as
shown below
Image by Author
t-values and P>|t|
The t-column provides the t-values corresponding to to each independent variables.

For example here Projects, People_managing and Salary all have different t-values
as well as different p-values associated with each variables. T-statistics are used to
calculate the p-values. Typically when p-value is less than 0.05, it indicates a strong
evidence against null hypothesis which states that the corresponding independent
variable has no effect on the dependent variable. P-value of 0.249 for Projects says
us that there is 24.9% chance that Projects variables has no effect on Salary. It seems
YearsExperience got 0 p-value indicating that the data for YearsExperience is
statistically significant since is is less than the critical limit (0.05). In this case, we
can reject the null hypothesis and say that YearsExperience data is significantly
controlling the Salary.
Years of Experience against Salary showing strong correlation
F-statistics
F-test provides a way to check all the independent variables all together if any of
those are related to the dependent variable. If Prob(F-statistic) is greater than 0.05,
there is no evidence of relationship between any of the independent variable with
the output. If it is less than 0.05, we can say that there is at least one variable which
is significantly related with the output. In our example, the p-value is less than 0.05
and therefore, one or more than one of the independent variable are related to
output variable Salary. We have seen previously that YearsExperience is significantly
related with Salary but others are not. Therefore, the F-test data supports the t-test
outcomes. However, there may be some cases when prob(F-statistic) may be greater
than 0.05 but one of the independent variable shows strong correlation. This is
because each t-test is carried out with different set of data whereas F-test checks the
combined effect including all variables globally.
Log-likelihood
The log-likelihood value is a measure for fit of the model with the given data. It is
useful when we compare two or more models. The higher the value of log-
likelihood, the better the model fits the given data. It can range from negative
infinity to positive infinity.
log-likelihood when all three variables are included
log-likelihood when only “Projects” is included
When all three independent variables are incorporated in the model, the log-
likelihood value is -310.21 which is higher than -334.95 when only Projects data is
included. This mean the first model fits the data better. It also goes hand in hand
with R-squared values as seen above.
AIC and BIC
AIC (stands for Akaike’s Information Criteria developed by Japanese statistician

Hirotugo Akaike) and BIC (stands for Bayesian Information Criteria) are also used as
criteria for model robustness. The goal is to minimize these values to get a better
model. I have another article where I have discussed on these topics.
Simple Stepwise and Weighted Regression Model

Stepwise and Weighted Regression
Model Stepwise and Weighted Regressiontowardsdatascience.com
Omnibus and Prob(Omnibus)
Omnibus test checks the normality of the residuals once the model is deployed. If
the value is zero, it means the residuals are perfectly normal. Here, in the example
prob(Omnibus) is 0.357 indicating that there is 35.7% chance that the residuals the
normally distributed. For a model to be robust, besides checking R-squared and
other rubrics, the residual distribution is also required to be normal ideally. In other
words, the residual should not follow any pattern when plotted against the fitted
values.
Skew and Kurtosis
Skew values tells us the skewness of the residual distribution. Normally distributed
variables have 0 skew values. Kurtosis is a measure of light-tailed or heavy-tailed
distribution compared to normal distribution. High kurtosis indicates the
distribution is too narrow and low kurtosis indicates the distribution is too flat. A
kurtosis value between -2 and +2 is good to prove normalcy.
Durbin-Watson
Durbin-Watson statistic provides a measure of autocorrelation in the residual. If the

residual values are autocorrelated, the model becomes biased and it is not expected.
This simply means that one value should not be depending on any of the previous
values. An ideal value for this test ranges from 0 to 4.
Jarque-Bera (JB) and Prob(JB)
Jarque-Bera (JB) and Prob(JB) is similar to Omni test measuring the normalcy of the
residuals.
Condition Number
High condition number indicates that there are possible multicollinearity present
in the dataset. If only one variable is used as predictor, this value is low and can be
ignored. We can proceed like stepwise regression and see if there is any
multicollinearity added when additional variables are included.
Conclusion
We have discussed all the summary parameters from statsmodel output. This will
useful for readers who are interested to check all the rubrics for a robust model.,
Most of the time, we look for R-squared value to make sure that the model explains
most of the variability but we have seen that there is much more than that.
Thanks for reading
Github page
Youtube Channel
Join Medium with my referral link - Md Sohel Mahmood

As a Medium member, a portion of your membership fee goes to
writers you read, and you get full access to every story…
mdsohel-mahmood.medium.com
Multiple Linearregression Simple Linear Regression Statsmodels Python
Statistics
Follow
Written by Md Sohel Mahmood

412 Followers · Writer for Towards Data Science
Data Science Enthusiast
More from Md Sohel Mahmood and Towards Data Science
Md Sohel Mahmood in Towards Data Science
Outlier Detection (Part 1)

IQR, Standard Deviation, Z-score and Modified Z-score
· 6 min read · May 5, 2022
109 4
Heiko Hotz in Towards Data Science
RAG vs Finetuning — Which Is the Best Tool to Boost Your LLM

Application?
The definitive guide for choosing the right method for your use case
· 19 min read · Aug 25
1.8K 16
Giuseppe Scalamogna in Towards Data Science
New ChatGPT Prompt Engineering Technique: Program Simulation

A potentially novel technique for turning a ChatGPT prompt into a mini-app.
9 min read · Sep 4
887 10
Md Sohel Mahmood in Towards Data Science
Factor Analysis of Mixed Data

Use of FAMD for data having continuous and categorical features
· 5 min read · Jul 12, 2021
32 1
See all from Md Sohel Mahmood
See all from Towards Data Science
Recommended from Medium
Erkan Hatipoğlu in Towards AI
Multivariate Linear Regression From Scratch

With Code in Python
12 min read · Mar 21
60
Maninder Singh
Understanding Categorical Correlations with Chi-Square Test and

Cramer’s V
In this, I explained about correlation(code) between categorical features which I Learned when
wanted to find the same in one of my…
9 min read · Jun 18
151
Lists
Predictive Modeling w/ Python

20 stories · 385 saves
Coding & Development

Practical Guides to Machine Learning

New_Reading_List
Yennhi95zz
Time Series in Machine Learning: Understanding and Applications

Time series analysis is one such method used to analyze and predict trends in time-dependent
data. In this blog post, we will explore what…
· 4 min read · Mar 31
203 3
Vivekawasthi
The Central Limit Therom & Making Estimates with the Confidence
intervals
This blog will cover the central limit theorem (CLT), allowing us to apply the concepts we
learned on the normal distribution to…
3 min read · Mar 24
Wendy Hu
Monte Carlo Simulation with Python

Introduction
5 min read · Apr 5
47
Juan Broglio
Gamma Regression vs Linear Regression (in Python)

General Linear Models and Gamma Regression
4 min read · Aug 9
See more recommendations

Simple Explanation of Statsmodel Linear Regression Model Summary

Uploaded by

Copyright:

Available Formats

Simple Explanation of Statsmodel Linear Regression Model Summary

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Simple Explanation of Statsmodel Linear Regression Model Summary

Uploaded by

Copyright:

Available Formats

9/17/23, 11:47 AM Simple Explanation of Statsmodel Linear Regression Model Summary | by Md Sohel Mahmood | Towards Data Science

Simple Explanation of Statsmodel Linear

Md Sohel Mahmood · Follow

Listen Share More

Typical model summary

I am going to explain all these parameters in the summary below.

“Salary” which is the only dependent variable in the data.

Model and Method

DF residuals and DF model

We have total 30 observation and 4 features. Out of 4 features, 3 features are

Covariance type is typically nonrobust which means there is no elimination of data

R-squared value is the coefficient of determination which indicates the percentage

Before moving to F-statistics, we need to understand the t-statistics first. T-statistics

coef and std err

t-values and P>|t|

The t-column provides the t-values corresponding to to each independent variables.

Years of Experience against Salary showing strong correlation

log-likelihood when all three variables are included

log-likelihood when only “Projects” is included

AIC and BIC

AIC (stands for Akaike’s Information Criteria developed by Japanese statistician

Simple Stepwise and Weighted Regression Model

Omnibus and Prob(Omnibus)

Skew and Kurtosis

Durbin-Watson statistic provides a measure of autocorrelation in the residual. If the

Jarque-Bera (JB) and Prob(JB)

Thanks for reading

Join Medium with my referral link - Md Sohel Mahmood

Multiple Linearregression Simple Linear Regression Statsmodels Python

Written by Md Sohel Mahmood

Data Science Enthusiast

More from Md Sohel Mahmood and Towards Data Science

Md Sohel Mahmood in Towards Data Science

Outlier Detection (Part 1)

· 6 min read · May 5, 2022

Heiko Hotz in Towards Data Science

RAG vs Finetuning — Which Is the Best Tool to Boost Your LLM

· 19 min read · Aug 25

Giuseppe Scalamogna in Towards Data Science

New ChatGPT Prompt Engineering Technique: Program Simulation

9 min read · Sep 4

Md Sohel Mahmood in Towards Data Science

Factor Analysis of Mixed Data

· 5 min read · Jul 12, 2021

See all from Md Sohel Mahmood

See all from Towards Data Science

Recommended from Medium

Erkan Hatipoğlu in Towards AI

Multivariate Linear Regression From Scratch

12 min read · Mar 21

Understanding Categorical Correlations with Chi-Square Test and

9 min read · Jun 18

Predictive Modeling w/ Python

Coding & Development

Practical Guides to Machine Learning

Time Series in Machine Learning: Understanding and Applications

· 4 min read · Mar 31

3 min read · Mar 24

Monte Carlo Simulation with Python

5 min read · Apr 5