Got it, let's work through this to get your bas.lm setup exactly how you want it — with the model containing only Predictor1/2/3 as the baseline (BF=1) at the top of your summary.

First, Fix the include.always Syntax #

Your current include.always has a small syntax issue: it shouldn’t include the response variable (Mean_Accuracy). The parameter only needs the predictors you want to force into every model. Correct it to:

include.always = ~ Predictor1 + Predictor2 + Predictor3

This locks in those three predictors, so your model space only has two possibilities: the baseline (no NewPredictor) and the extended model (baseline + NewPredictor).

Understand How bestmodel Maps to Your Predictors #

The bestmodel parameter is a binary vector where each element corresponds to a predictor in your main formula (Mean_Accuracy ~ .). The order matches the column order of your predictors in the data frame:

1. Predictor1
2. Predictor2
3. Predictor3
4. NewPredictor

Since you want to start sampling from your baseline model (exclude NewPredictor), set bestmodel to c(1, 1, 1, 0) — the first three 1s enforce the forced predictors, and the 0 excludes NewPredictor.

Full Corrected Code #

Here’s the updated code with all fixes:

# Your data (for reference)
df <- data.frame(
  Mean_Accuracy = c(0.73, 0.50, 0.86),
  Predictor1 = c(3, 4, 1),
  Predictor2 = c(7, 5, 3),
  Predictor3 = c(6, 1, 4),
  NewPredictor = c(8, 5, 1)
)

# Run the model with correct parameters
Accuracy <- bas.lm(
  Mean_Accuracy ~ ., 
  data = df, 
  method = "BAS", 
  bestmodel = c(1, 1, 1, 0),  # Initialize with baseline model
  prior = "JZS", 
  include.always = ~ Predictor1 + Predictor2 + Predictor3,  # Force these predictors
  modelprior = uniform()
)

# Check the summary
summary(Accuracy)

Why This Works #

• The corrected include.always ensures Predictor1/2/3 are in every model, so only NewPredictor is optional.
• Setting bestmodel = c(1,1,1,0) starts the MCMC sampling from your baseline model. With only two possible models, this will make the baseline appear first in the summary (with BF=1, as it’s the reference point).
• The second model (with NewPredictor) will show its Bayes Factor relative to this baseline, which is exactly what you’re aiming for.

Quick Note #

If for some reason the baseline still doesn’t show up first, you can manually reorder the models in the output, but with such a small model space, the bestmodel initialization should prioritize it automatically.

内容的提问来源于stack exchange，提问作者Agata