BUG: emmeans backend doesn't parse well factors with "-"

[DominiqueMakowski]

df <- data.frame(
  x = rnorm(1000),
  y = rep_len(c("A - High", "A - Low", "B - High", "B - Low"), 1000)
)

model <- lm(x ~ y, data=df)
modelbased::estimate_contrasts(model, contrast = "y", backend = "marginaleffects")  # Works
#> Marginal Contrasts Analysis
#> 
#> Level1   | Level2   | Difference |   SE |        95% CI | t(996) |     p
#> ------------------------------------------------------------------------
#> A - Low  | A - High |       0.02 | 0.09 | [-0.16, 0.19] |   0.17 | 0.863
#> B - High | A - High |       0.01 | 0.09 | [-0.16, 0.18] |   0.13 | 0.898
#> B - Low  | A - High |      -0.05 | 0.09 | [-0.22, 0.13] |  -0.53 | 0.597
#> B - High | A - Low  |  -3.95e-03 | 0.09 | [-0.18, 0.17] |  -0.05 | 0.964
#> B - Low  | A - Low  |      -0.06 | 0.09 | [-0.23, 0.11] |  -0.70 | 0.483
#> B - Low  | B - High |      -0.06 | 0.09 | [-0.23, 0.11] |  -0.66 | 0.511
#> 
#> Variable predicted: x
#> Predictors contrasted: y
#> p-values are uncorrected.
modelbased::estimate_contrasts(model, contrast = "y", backend = "emmeans")  
#> Marginal Contrasts Analysis
#> 
#> Level | X2    | X3 | X4    | Difference |        95% CI |   SE | t(996) |     p
#> -------------------------------------------------------------------------------
#> (A    | Low)  | (A | High) |       0.02 | [-0.16, 0.19] | 0.09 |   0.17 | 0.863
#> (B    | High) | (A | High) |       0.01 | [-0.16, 0.18] | 0.09 |   0.13 | 0.898
#> (B    | High) | (A | Low)  |  -3.95e-03 | [-0.18, 0.17] | 0.09 |  -0.05 | 0.964
#> (B    | Low)  | (A | High) |      -0.05 | [-0.22, 0.13] | 0.09 |  -0.53 | 0.597
#> (B    | Low)  | (A | Low)  |      -0.06 | [-0.23, 0.11] | 0.09 |  -0.70 | 0.483
#> (B    | Low)  | (B | High) |      -0.06 | [-0.23, 0.11] | 0.09 |  -0.66 | 0.511
#> 
#> Variable predicted: x
#> Predictors contrasted: y
#> p-values are uncorrected.

^{Created on 2026-07-28 with reprex v2.1.1}

[strengejacke]

See code starting from here, which I wrote for marginaleffects backend to solve (or better' workaround) this problem:

modelbased/R/format.R

Line 400 in e12e4a9

# create replacement vector

I didn't touch your emmeans implementation, because I had no motivation to look closer into the code 😜

[DominiqueMakowski]

Poor emmeans unloved by daniel 🦆