Reporting your modelled effects

1 Reporting your modelled effects

In the last chapter of these notes, we discussed testing your hypothesis.

Remember that testing your hypothesis is looking for evidence of effects of your predictors on your response. By testing your hypothesis, you determined that there was evidence that the effects of your predictors on your response are non-zero (Popovic et al. 2024).

These effects are presented as estimates of the coefficients in your model (e.g. a slope) which are called parameters once they have been estimated.

For categorical predictors, the coefficient represents how much your response changes (increases or decreases) when you change from one level of the category to another.

For numeric predictors, the coefficient represents how much your response changes (increases or decreases) when you increase your predictor by one unit. More on this below!

Here you will gather the information to communicate visually and quantitatively the magnitude and direction of your effects and determine (in some cases) where effects are similar or different from one another, answering questions like:

  • how much does a change (effect) in your predictor change your response?

  • is that change (effect) positive (your response increases), or negative (your response decreases)?

  • is that change (effect) similar across levels of your categorical predictor?

You can report your modelled effects in a number of ways1:

1.1 Visualizing your model effects

Here, you will make plots of your model effects including the effects themselves, uncertainty around the effects and your observations. These plots will help communicate your modelled effects as well as how well your model fits your observations. This is a great way to communicate all types of model structures (e.g. error distribution and shape assumptions).

1.2 Giving examples of your modelled effects

Here, you will use your model to estimate the value of your response (with uncertainty) to illustrate how the response changes when your predictors change. This is a great way to communicate all types of model structures (e.g. error distribution and shape assumptions).

1.3 Quantifying your model effects

Here, you will report (in numbers) the magnitude and direction of your modelled effects along with the uncertainty. You will also look for evidence of whether modelled effects of a categorical predictor differ across all levels. Exactly how you do that will depend on the structure of your model (e.g. the error distribution and shape assumptions). We will go through a number of examples here.

Data Skills Portfolio Program CC BY-NC-SA 4.0

2 Back to Reporting main page

Popovic, Gordana, Tanya Jane Mason, Szymon Marian Drobniak, Tiago André Marques, Joanne Potts, Rocío Joo, Res Altwegg, et al. 2024. “Four Principles for Improved Statistical Ecology.” Methods in Ecology and Evolution 15 (2): 266–81. https://doi.org/10.1111/2041-210x.14270.

Footnotes

  1. with some options being more or less helpful depending on your model structure↩︎