correlation() and cor_test() handle factors differently when using method = spearman

[axrhart]

Dear authors,

a student pointed out that the following uses of correlation() and cor_test() return different results.

set.seed(1337)

data.frame(
  a = sample(1:5, 10, replace = T),
  b = factor(sample(c("s", "m", "l"), 10, , replace = T), 
             levels = c("s", "m", "l"), 
             ordered = T)
) -> df

correlation::correlation(df, method = "spearman")

# # Correlation Matrix (spearman-method)
# 
# Parameter1 | Parameter2 |   rho |         95% CI |      S |     p
# -----------------------------------------------------------------
#   a          |        b.s | -0.49 | [-0.86,  0.22] | 245.72 | 0.668
# a          |        b.m | -0.11 | [-0.70,  0.57] | 182.97 | 0.765
# a          |        b.l |  0.51 | [-0.20,  0.87] |  81.12 | 0.668
# b.s        |        b.m | -0.41 | [-0.83,  0.32] | 232.36 | 0.725
# b.s        |        b.l | -0.41 | [-0.83,  0.32] | 232.36 | 0.725
# b.m        |        b.l | -0.67 | [-0.92, -0.04] | 275.00 | 0.212
# 
# p-value adjustment method: Holm (1979)
# Observations: 10
# Warning message:
#   It seems like there is not enough continuous variables in your data. Maybe you want to include the factors? 
#   We're setting `include_factors=TRUE` for you. 

correlation::cor_test(df, "a", "b", method = "spearman")

# Parameter1 | Parameter2 |  rho |        95% CI |     S |     p
# --------------------------------------------------------------
# a          |          b | 0.59 | [-0.08, 0.89] | 67.64 | 0.073
# 
# Observations: 10

I wondered whether this behavior is intentional?
If yes, which one would you consider the correct approach to compute rho between an ordered factor and a numeric variable?

Best,
Alexander

[bwiernik]

The proper method for ordered factors (but not unordered factors) should probably be the cor_test behavior of converting the ordered levels to integers (i.e., ranks).

[axrhart]

Thank you. Then we will stick to using cor_test() in our materials. 👍