Dichotomize or not? Logistic vs Linear

  • A
  • Thread starter FallenApple
  • Start date
  • Tags
    Linear
In summary, the conversation discussed the use of linear regression and logistic regression to determine the association between a continuous variable, strength, and another continuous variable, height. The advantage of each method was debated, with one providing an estimate of the odds of being tall and the other providing an estimate of height. The confusion arose when trying to determine the association between strength and weight. Ultimately, either method can provide an association between the two variables.
  • #1
FallenApple
566
61
Say I want to see if a continuous variable, say strength, is associated with another continuous variable, say height.

Clearly I can just use linear regression with height as the response variable and strength as the predictor variable.

But I can also split height into a tall category and a not tall category. And then do logistic regression with the log odds of height as the response. So what is the advantage of one way vs the other?
 
Physics news on Phys.org
  • #2
Do you want an estimate of the odds of being tall or do you want an estimate of the height?
 
  • #3
Dale said:
Do you want an estimate of the odds of being tall or do you want an estimate of the height?
I guess that's the confusing part. What should I do if I want to know if strength is in any way associated with weight generally?

If I do linear regression, I can interpret the change in mean height for unit increases in strength. But for the odds, I can do the change in odds of being tall for every unit increase in strength. But either way, wouldn't either give an association? If on average, height increases when strength increases, then I would be surprised if the odds didn't increase as well.
 

1. What is the difference between dichotomize and not dichotomize in logistic and linear regression?

Dichotomization refers to the process of converting a continuous variable into a binary variable, while not dichotomizing means keeping the variable as continuous. Logistic regression is typically used for binary outcomes and requires dichotomized predictors, while linear regression can handle both continuous and binary predictors.

2. When should I dichotomize my predictors in logistic regression?

Dichotomization should be done when the underlying relationship between the predictor and outcome is expected to be non-linear. In logistic regression, the relationship between the predictor and outcome is assumed to be logit-linear, so dichotomizing can improve the model's fit in some cases.

3. What are the potential drawbacks of dichotomizing predictors in logistic regression?

Dichotomizing predictors can lead to loss of information and decrease the power of the analysis. It can also introduce bias and make the interpretation of the results more difficult.

4. Can I use both dichotomized and continuous predictors in the same logistic regression model?

Yes, it is possible to include both dichotomized and continuous predictors in a logistic regression model. However, this can result in a more complex model and should be done with caution.

5. Is it always better to use logistic regression over linear regression for binary outcomes?

No, it is not always better to use logistic regression over linear regression for binary outcomes. If the relationship between the predictor and outcome is expected to be linear, then linear regression may be a more appropriate choice. It is important to consider the underlying assumptions and characteristics of the data before deciding on the appropriate regression model.

Similar threads

I Exploring OLS and GLM Models: Understanding the Link Function and Coefficients
    • Set Theory, Logic, Probability, Statistics
Replies
3
Views
841
I How to Express Non-regular Prior Distributions by Mathematical Formula
    • Set Theory, Logic, Probability, Statistics
Replies
2
Views
353
I Time Series Models vs Models Used for Cross-sectional Data
    • Set Theory, Logic, Probability, Statistics
Replies
14
Views
256
I Expected coefficient change from simple to multiple linear regression
    • Set Theory, Logic, Probability, Statistics
Replies
4
Views
1K
I Coefficient sign flip in linear regression with correlated predictors
    • Set Theory, Logic, Probability, Statistics
Replies
13
Views
2K
I Linear regression, feature scaling, and regression coefficients
    • Set Theory, Logic, Probability, Statistics
Replies
8
Views
1K
A Continuous output: logistic vs linear regression
    • Set Theory, Logic, Probability, Statistics
Replies
4
Views
1K
A How Do Generalized Linear Models Extend Beyond Standard Linear Regression?
    • Set Theory, Logic, Probability, Statistics
Replies
9
Views
2K
B Comparing Approaches: Linear Regression of Y on X vs X on Y
    • Set Theory, Logic, Probability, Statistics
2
Replies
64
Views
3K
A Observing interactions with plots using est. coeff.
    • Set Theory, Logic, Probability, Statistics
Replies
4
Views
1K

Hot Threads

Recent Insights

Back
Top