vlt.stats.power.simulate_lme_data_hierarchical

  SIMULATE_LME_DATA_HIERARCHICAL - (Helper) Generates data using the 'hierarchical' method.

    Methodology:
    This is a sophisticated non-parametric method that respects the nested
    structure of a mixed-effects model. It simulates the null hypothesis by
    shuffling the outcome variable *within* each random effect group. This
    breaks the association between the fixed effects and the outcome, while
    correctly preserving the dependencies among observations in the same group.

    Process:
    1.  For each group (e.g., for each manufacturer 'Mfg'), it takes all the
        observed Y values.
    2.  It shuffles these Y values *only within that group*.
    3.  This is repeated independently for every group.
    4.  The result is a simulated null dataset where the within-group data
        structure is maintained, but the relationship between Y and the fixed
        predictors (like 'Model_Year') is randomized.
    5.  Injects the `effect_size` to create the alternative hypothesis dataset.

    Assumptions:
    -   Assumes "exchangeability" of observations *within* each group.

    Pros/Cons:
    +   Robust to violations of normality.
    +   Theoretically sound for mixed models as it preserves the random-effects
        structure. It is generally the preferred non-parametric method.
    -   May be less effective if some groups have very few observations to shuffle.