What does it mean when I add a new variable to my linear model and the R^2 stays the same?
$begingroup$
I'm inclined to think that the new variable is not correlated to the response. But could the new variable be correlated to another variable in the model?
linear-model r-squared
asked yesterday
Chance113Chance113
362
$endgroup$
|
show 2 more comments
$begingroup$
I'm inclined to think that the new variable is not correlated to the response. But could the new variable be correlated to another variable in the model?
linear-model r-squared
asked yesterday
Chance113Chance113
362
$endgroup$
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
5
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
5
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago
|
show 2 more comments
$begingroup$
I'm inclined to think that the new variable is not correlated to the response. But could the new variable be correlated to another variable in the model?
linear-model r-squared
asked yesterday
Chance113Chance113
362
$endgroup$
I'm inclined to think that the new variable is not correlated to the response. But could the new variable be correlated to another variable in the model?
linear-model r-squared
linear-model r-squared
asked yesterday
Chance113Chance113
362
asked yesterday
Chance113Chance113
362
asked yesterday
Chance113Chance113
362
asked yesterday
Chance113Chance113
362
asked yesterday
Chance113Chance113
362
362
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
5
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
5
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago
|
show 2 more comments
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
5
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
5
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
5
5
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
5
5
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago
|
show 2 more comments
2 Answers
2
active
oldest
votes
$begingroup$
Seeing little to no change in $R^2$ when you add a variable to a linear model means that the variable has little to no additional explanatory power to the response over what is already in your model. As you note, this can be either because it tells you almost nothing about the response or it explains the same variation in the response as the variables already in the model.
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
$endgroup$
add a comment |
$begingroup$
As others have alluded, seeing no change in $R^2$ when you add a variable to your regression is unusual. In finite samples, this should only happen when your new variable is a linear combination of variables already present. In this case, most standard regression routines simply exclude that variable from the regression, and your $R^2$ will remain unchanged because the model was effectively unchanged.
As you notice, this does not mean the variable is unimportant, but rather that you are unable to distinguish its effect from that of the other variables in your model.
More broadly however, I (and many here at Cross Validated) would caution against using R^2 for model selection and interpretation. What I've discussed above is how the $R^2$ could not change and the variable still be important. Worse yet, the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable. Broadly, using $R^2$ for model selection fell out of favor in the 70s, when it was dropped in favor of AIC (and its contemporaries). Today -- a typical statistician would recommend using cross validation (see the site name) for your model selection.
In general, adding a variable increases $R^2$ -- so using $R^2$ to determine a variables importance is a bit of a wild goose chase. Even when trying to understand simple situations you will end up with a completely absurd collection of variables.
answered yesterday
user5957401user5957401
29727
$endgroup$
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
add a comment |
Your Answer
StackExchange.ifUsing("editor", function () {
return StackExchange.using("mathjaxEditing", function () {
StackExchange.MarkdownEditor.creationCallbacks.add(function (editor, postfix) {
StackExchange.mathjaxEditing.prepareWmdForMathJax(editor, postfix, [["$", "$"], ["\\(","\\)"]]);
});
});
}, "mathjax-editing");
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "65"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});
function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: false,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: null,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});
}
});
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstats.stackexchange.com%2fquestions%2f396220%2fwhat-does-it-mean-when-i-add-a-new-variable-to-my-linear-model-and-the-r2-stays%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
2 Answers
2
active
oldest
votes
2 Answers
2
active
oldest
votes
active
oldest
votes
active
oldest
votes
$begingroup$
Seeing little to no change in $R^2$ when you add a variable to a linear model means that the variable has little to no additional explanatory power to the response over what is already in your model. As you note, this can be either because it tells you almost nothing about the response or it explains the same variation in the response as the variables already in the model.
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
$endgroup$
add a comment |
$begingroup$
Seeing little to no change in $R^2$ when you add a variable to a linear model means that the variable has little to no additional explanatory power to the response over what is already in your model. As you note, this can be either because it tells you almost nothing about the response or it explains the same variation in the response as the variables already in the model.
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
$endgroup$
add a comment |
$begingroup$
Seeing little to no change in $R^2$ when you add a variable to a linear model means that the variable has little to no additional explanatory power to the response over what is already in your model. As you note, this can be either because it tells you almost nothing about the response or it explains the same variation in the response as the variables already in the model.
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
$endgroup$
Seeing little to no change in $R^2$ when you add a variable to a linear model means that the variable has little to no additional explanatory power to the response over what is already in your model. As you note, this can be either because it tells you almost nothing about the response or it explains the same variation in the response as the variables already in the model.
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
answered yesterday
TrynnaDoStatTrynnaDoStat
5,56211335
5,56211335
add a comment |
add a comment |
$begingroup$
As others have alluded, seeing no change in $R^2$ when you add a variable to your regression is unusual. In finite samples, this should only happen when your new variable is a linear combination of variables already present. In this case, most standard regression routines simply exclude that variable from the regression, and your $R^2$ will remain unchanged because the model was effectively unchanged.
As you notice, this does not mean the variable is unimportant, but rather that you are unable to distinguish its effect from that of the other variables in your model.
More broadly however, I (and many here at Cross Validated) would caution against using R^2 for model selection and interpretation. What I've discussed above is how the $R^2$ could not change and the variable still be important. Worse yet, the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable. Broadly, using $R^2$ for model selection fell out of favor in the 70s, when it was dropped in favor of AIC (and its contemporaries). Today -- a typical statistician would recommend using cross validation (see the site name) for your model selection.
In general, adding a variable increases $R^2$ -- so using $R^2$ to determine a variables importance is a bit of a wild goose chase. Even when trying to understand simple situations you will end up with a completely absurd collection of variables.
answered yesterday
user5957401user5957401
29727
$endgroup$
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
add a comment |
$begingroup$
As others have alluded, seeing no change in $R^2$ when you add a variable to your regression is unusual. In finite samples, this should only happen when your new variable is a linear combination of variables already present. In this case, most standard regression routines simply exclude that variable from the regression, and your $R^2$ will remain unchanged because the model was effectively unchanged.
As you notice, this does not mean the variable is unimportant, but rather that you are unable to distinguish its effect from that of the other variables in your model.
More broadly however, I (and many here at Cross Validated) would caution against using R^2 for model selection and interpretation. What I've discussed above is how the $R^2$ could not change and the variable still be important. Worse yet, the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable. Broadly, using $R^2$ for model selection fell out of favor in the 70s, when it was dropped in favor of AIC (and its contemporaries). Today -- a typical statistician would recommend using cross validation (see the site name) for your model selection.
In general, adding a variable increases $R^2$ -- so using $R^2$ to determine a variables importance is a bit of a wild goose chase. Even when trying to understand simple situations you will end up with a completely absurd collection of variables.
answered yesterday
user5957401user5957401
29727
$endgroup$
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
add a comment |
$begingroup$
As others have alluded, seeing no change in $R^2$ when you add a variable to your regression is unusual. In finite samples, this should only happen when your new variable is a linear combination of variables already present. In this case, most standard regression routines simply exclude that variable from the regression, and your $R^2$ will remain unchanged because the model was effectively unchanged.
As you notice, this does not mean the variable is unimportant, but rather that you are unable to distinguish its effect from that of the other variables in your model.
More broadly however, I (and many here at Cross Validated) would caution against using R^2 for model selection and interpretation. What I've discussed above is how the $R^2$ could not change and the variable still be important. Worse yet, the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable. Broadly, using $R^2$ for model selection fell out of favor in the 70s, when it was dropped in favor of AIC (and its contemporaries). Today -- a typical statistician would recommend using cross validation (see the site name) for your model selection.
In general, adding a variable increases $R^2$ -- so using $R^2$ to determine a variables importance is a bit of a wild goose chase. Even when trying to understand simple situations you will end up with a completely absurd collection of variables.
answered yesterday
user5957401user5957401
29727
$endgroup$
As others have alluded, seeing no change in $R^2$ when you add a variable to your regression is unusual. In finite samples, this should only happen when your new variable is a linear combination of variables already present. In this case, most standard regression routines simply exclude that variable from the regression, and your $R^2$ will remain unchanged because the model was effectively unchanged.
As you notice, this does not mean the variable is unimportant, but rather that you are unable to distinguish its effect from that of the other variables in your model.
More broadly however, I (and many here at Cross Validated) would caution against using R^2 for model selection and interpretation. What I've discussed above is how the $R^2$ could not change and the variable still be important. Worse yet, the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable. Broadly, using $R^2$ for model selection fell out of favor in the 70s, when it was dropped in favor of AIC (and its contemporaries). Today -- a typical statistician would recommend using cross validation (see the site name) for your model selection.
In general, adding a variable increases $R^2$ -- so using $R^2$ to determine a variables importance is a bit of a wild goose chase. Even when trying to understand simple situations you will end up with a completely absurd collection of variables.
answered yesterday
user5957401user5957401
29727
answered yesterday
user5957401user5957401
29727
answered yesterday
user5957401user5957401
29727
answered yesterday
user5957401user5957401
29727
29727
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
add a comment |
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
$begingroup$
Could you elaborate on the $R^2$ could change somewhat (or even dramatically) when you include an irrelevant variable, specifically on the case of a dramatical change? In which sense would the variable then be irrelevant?
$endgroup$
– Richard Hardy
yesterday
add a comment |
Thanks for contributing an answer to Cross Validated!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
Use MathJax to format equations. MathJax reference.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstats.stackexchange.com%2fquestions%2f396220%2fwhat-does-it-mean-when-i-add-a-new-variable-to-my-linear-model-and-the-r2-stays%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
$begingroup$
It depends, could you provide us with some reduced data lines or output from your linear models. Without more information it's hard to assist you
$endgroup$
– OliverFishCode
yesterday
5
$begingroup$
It shouldn't stay exactly the same unless it is perfectly orthogonal to your response, or is a linear combination of the variables already included. It may be that the change is smaller than the number of decimal places displayed.
$endgroup$
– gung♦
yesterday
5
$begingroup$
@gung What you can infer is that the new variable is orthogonal to the response modulo the subspace generated by the other variables. That's more general than the two options you mention.
$endgroup$
– whuber♦
yesterday
$begingroup$
@whuber, yes, I suppose so.
$endgroup$
– gung♦
yesterday
$begingroup$
Test your variables for multicollinearity en.wikipedia.org/wiki/Multicollinearity probably some features are linearly connected. Use caret package and vif() in R sthda.com/english/articles/39-regression-model-diagnostics/…
$endgroup$
– Tom Zinger
23 hours ago