I dont know how to correctly present a logistic regression model in expressions or formula in a manuscript or a report, especially with a multiple-level categorical variable. For instance, I have a 3-level treatments (treatment) as the explanatory variable: control, low, and high. The outcome (Y) is alive or dead. Can someone suggest how to write down the formula? Is there a standard formula or an easy understandable formula, especially for non-statistical readers (biology or medicine)? I am bit afraid that if I present the model using matrix format, the readers would not get the idea that there are two values of coefficient beta. But for me, it would be nice to see more forms of presentations.

This is what I can think of:

Y_i ~ Binomial(1,p_i)

logit(p_i) = intercept + beta_k*treatment_i

where i indicates the ith sample. For beta_k, k=low when the ith sample has treatment low; and k=high when the ith sample has treatment high.

Thanks very much

  • 1
    $\begingroup$ Notation tends to differ a lot by (sub-(sub-(sub-)))discipline. What is your intended audience? $\endgroup$ Apr 16, 2015 at 7:13
  • $\begingroup$ @Maarten Buis, it is for non-statistical readers. But I would like to see more variations of the presentations. Thanks $\endgroup$ Apr 16, 2015 at 7:20
  • $\begingroup$ Non-statistical only tells us who they are not, so that does not narrow it down a lot. Are these academics or not? If yes, what discipline? If not, who are they (policy people, high school students, ...)? $\endgroup$ Apr 16, 2015 at 7:31
  • $\begingroup$ @Maarten Buis, they are academics in medical or biological field. $\endgroup$ Apr 16, 2015 at 8:06

2 Answers 2


If your audience is non-statistical then your first line declaring the distribution of $Y_i$ to be binomial will just tend to confuse more than help. So I would just leave that out.

With only three treatments and a non-technical audience I don't see the added value of trying anything fancy. Instead I would just mention those two indicator (dummy) variables directly.

Since your audience is from the bio-medical fields, they tend to be familiar with Odds, so you could formulate it in those terms:

$\ln(odds(Y_i=dead|x_i)) = \beta_0 + \beta_1 low_i + \beta_2 high_i $

You could do this in terms of the probability:

$\ln\left(\frac{p(Y_i=dead|x_i)}{1-p(Y_i=dead|x_i)}\right) = \beta_0 + \beta_1 low_i + \beta_2 high_i $


$p(Y_i=dead|x_i) = \frac{\exp(\beta_0 + \beta_1 low_i + \beta_2 high_i)}{1+\exp(\beta_0 + \beta_1 low_i + \beta_2 high_i)}$

  • $\begingroup$ Thanks @Maarten Buis. Indeed it's more straightforward to formulate directly as odds. However, I am bit confused, the left side of the formula you used notation _i, but it is not in the right side of the formula anymore. Do you think it might be confusing for readers? $\endgroup$ Apr 17, 2015 at 9:47
  • $\begingroup$ I have edited my answer. I usually remove the i subscript alltogether (less symbols in your formula means less things to explain so you can focus on the important stuff and less things that can be misunderstood), but you can add them to $high$ and $low$ if you like them in. $\endgroup$ Apr 17, 2015 at 10:08

When you have categorical variables, lets say X takes values A, B, C, D then the first level called baseline and it's zero.

So you have the same fomula log(p/(1-p)) = beta_{1}*X_{B} + beta_{2}*X_{C} + beta_{3}*X_{C}

where X_{i} are idicators and takes 1 when i is true (when you are in level b or c or d) and 0 otherwise.

Hope this help you.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.