Multilevel correlation vs. linear mixed model

[pedfig]

I was trying to compute the multilevel correlations using the package and a linear mixed model. Still, I don’t seem to be able to calculate precisely the values provided by the correlation package. Can you give me information on why this is the case? Below is an example. Thank you!

data %>% select(ID, var1, var2) %>%
    correlation(partial = FALSE, multilevel = TRUE, p_adjust = "none")

Parameter1 | Parameter2 |     r |             CI | t(70) |      p
-----------------------------------------------------------------
var1             |      var2     | -0.29 | [-0.49, -0.07] | -2.57 | 0.012*


example <- lmerTest::lmer(var1 ~ var2 + (1|ID), data = data)

parameters::model_parameters(example, effects = "fixed", ci_method = "satterthwaite")

Parameter   | Coefficient |    SE |           95% CI |     t |    df |      p
-----------------------------------------------------------------------------
(Intercept) |      424.12 | 10.58 | [401.75, 446.48] | 40.09 | 16.54 | < .001
var2       |       -9.43 |  3.45 | [-16.32,  -2.54] | -2.74 | 61.34 | 0.008 

t_to_r(-2.74, df_error = 61.34)

r     |         95% CI
----------------------
-0.33 | [-0.52, -0.09]

[DominiqueMakowski]

Sorry I missed that.

As you can see, the operation is taken care of here:

correlation/R/cor_test.R

Lines 151 to 154 in a5444ca

	if (isTRUE(partial)) {
	data[[x]] <- datawizard::adjust(data[names(data) != y], multilevel = multilevel, bayesian = partial_bayesian)[[x]]
	data[[y]] <- datawizard::adjust(data[names(data) != x], multilevel = multilevel, bayesian = partial_bayesian)[[y]]
	}

which calls adjust with multilevel set to TRUE. That happens in cor_test(). However in correlation() there's also this step:

correlation/R/correlation.R

Line 392 in a5444ca

if (convert_back_to_r) out <- pcor_to_cor(pcor = out) # Revert back to r if needed.

Maybe that's the cause? To be honest I'm not sure, one would need to reproduce the different steps manually

[pedfig]

Thank you!