tests/testthat/test_custom_effect_sizes.R

In papaja: Prepare American Psychological Association Journal Articles with R Markdown

test_that(
  "Custom effect sizes for ANOVA"
  , {
    set.seed(49873)
    ges <- getOption("papaja.estimate_anova")

    # Treat as between-subjects design, use Type-1 sums of squares
    npk$id <- seq_len(nrow(npk))
    aov_out <- afex::aov_4(
      formula = yield ~ N * P + (1|id)
      , data = npk
      , include_aov = TRUE
    )$aov
    summary_aov <- summary(aov_out)



    # 'aov/lm' class
    apa_with_function <- apa_print(aov_out)
    apa_with_df <- apa_print(aov_out, es = ges(aov_out))
    apa_summary_aov <- apa_print(summary_aov, es = ges)

    expect_apa_results(
      apa_with_function
      , labels = list(
        term = "Effect"
        , estimate = "$\\hat{\\eta}^2_G$"
        , conf.int = "90\\% CI"
        , statistic = "$F$"
        , df = "$\\mathit{df}$"
        , df.residual ="$\\mathit{df}_{\\mathrm{res}}$"
        , p.value = "$p$"
      )
    )

    expect_identical(
      !!names(apa_with_df$estimate)
      , !!c("N", "P", "N_P")
    )
    expect_identical(
      !!apa_with_df$estimate$N
      , "$\\hat{\\eta}^2_G = .224$, 90\\% CI $[.017, .460]$"
    )

    expect_identical(
      apa_with_function
      , apa_with_df
    )

    expect_identical(
      object = apa_summary_aov
      , expected = apa_with_df
    )


    # Treat as within-subjects design, use Type-3 sums of squares
    # 'afex_aov' class
    afex_out <- afex::aov_4(yield ~ (N * P | block), data = npk, include_aov = TRUE)

    afex_with_intercept <- apa_print(afex_out, es = ges, intercept = TRUE)
    afex_without_intercept <- expect_warning(
      apa_print(afex_out, es = ges(afex_out, include_intercept = FALSE), intercept = TRUE)
      , regexp = "Custom effect sizes were not available for some model terms. These have been dropped from the output object."
    )

    # aovlist/listof (be careful: within subjects, Type-1 sums of squares)
    aovlist_out    <- apa_print(afex_out$aov, es = ges)
    aovlist_out_df <- apa_print(afex_out$aov, es = ges(afex_out$aov))

    expect_identical(
      aovlist_out
      , aovlist_out_df
    )

    # Anova.mlm (Type-3 again, but only partial eta-squared available)
    Anova.mlm_out    <- apa_print(afex_out$Anova) # expect_warning(): seems like warnings came from within the effectsize package
    Anova.mlm_out_df <- apa_print(afex_out$Anova, es = ges(afex_out$Anova))# expect_warning(): seems like warnings came from within the effectsize package

    expect_identical(
      Anova.mlm_out
      , Anova.mlm_out_df
    )

    # anova from car::Anova, between Ss
    afex_between <- afex::aov_4(formula = yield ~ N*P + (1|id), data = npk)

    expect_identical(
      apa_print(afex_between$Anova)
      , apa_print(afex_between)
    )

    # anova from anova.lm()
    # between Ss, Type-1 SSs
    stats_anova_lm <- anova(afex_between$lm)

    expect_identical(
      apa_print(stats_anova_lm)
      , apa_with_function
    )

    # Non-supported output from summary()
    # summary.Anova.mlm
    expect_error(
      apa_print(summary(afex_out$Anova), es = ges)
      , regexp = "Cannot apply custom effect-size function to this class of object."
    )

    skip_on_cran()
    expect_identical(
      !!apa_with_df$estimate$P
      , "$\\hat{\\eta}^2_G = .013$, 90\\% CI $[.000, .181]$"
    )
    expect_identical(
      !!apa_with_df$estimate$N_P
      , "$\\hat{\\eta}^2_G = .031$, 90\\% CI $[.000, .230]$"
    )
  }
)

test_that(
  "Calculation of generalized eta squared"
  , {
    library("afex")
    library("effectsize")

    suppressMessages(set_sum_contrasts())
    data(obk.long)

    afex_aov_between <- aov_ez(
      "id"
      , "value"
      , obk.long
      , between = c("treatment", "gender")
      , observed = "gender"
      , fun_aggregate = mean
    )

    apa_afex <- apa_print(afex_aov_between)

    expect_identical(
      apa_afex$table$estimate
      , structure(
        c(".289", ".189", ".295")
        , label = "$\\hat{\\eta}^2_G$"
        , class = c("tiny_labelled", "character")
      )
    )

    afex_aov_between2 <- aov_ez(
      "id"
      , "value"
      , obk.long
      , between = c("treatment", "gender")
      , observed = "treatment"
      , fun_aggregate = mean
    )

    expect_warning(
      apa_afex2 <- apa_print(afex_aov_between, observed = "treatment")
    )
    expect_identical(apa_print(afex_aov_between2), apa_afex2)
  }
)