Cofad User Guide

Contents

• Citation
• Introduction
• Installation
  • Docker
• Using cofad
  • Graphical user interface
  • Between-subjects designs
  • Within-subjects designs
    • Using participant L or r scores
  • Mixed designs
• Comparing two hypotheses
• Aggregated data
• Testing
• Browser-only webR version
• Issues and support
• Contributing
• Acknowledgments
• References

Citation

If you use cofad and find it useful, please cite both the current tutorial and the software paper:

@article{henninger2025cofad,
  author = {Henninger, Mirka and Malejka, Simone and Titz, Johannes},
  title = {Contrast analysis for competing hypotheses: A tutorial using the {R} package cofad},
  journal = {Behavior Research Methods},
  year = {2025},
  volume = {57},
  pages = {326},
  doi = {10.3758/s13428-025-02833-w}
}

@article{titz2021cofad,
  author = {Titz, Johannes and Burkhardt, Markus},
  title = {cofad: An {R} package and {Shiny} app for contrast analysis},
  journal = {Journal of Open Source Software},
  year = {2021},
  volume = {6},
  number = {67},
  pages = {3822},
  doi = {10.21105/joss.03822}
}

Introduction

Cofad is an R package for conducting COntrast analysis in FActorial Designs. If contrast analysis were to win an award, it might be for the most underestimated and underused statistical technique. This is unfortunate, because contrast analysis is at least as informative as ANOVA—and often considerably more so. Instead of testing an unspecific omnibus hypothesis such as “there are differences somewhere,” contrast analysis allows you to formulate and test precise, numerically specified hypotheses about the effects of interest. In doing so, it shifts the emphasis from merely asking whether an effect exists to evaluating its expected pattern and magnitude.

This focus on effects is reflected in two key ways:

1. Between-subjects and mixed analyses provide three distinct effect-size measures: \(r_\mathrm{es}\), \(r_\mathrm{contrast}\), and \(r_\mathrm{alerting}\). Pure within-subjects analyses instead report \(r_\mathrm{contrast}\) and \(g_\mathrm{contrast}\), because the other two measures rely on a between-group partition.
2. These effect sizes capture different aspects of how closely the observed data conform to the specified contrast.

Cofad also allows you to compare two competing hypotheses directly (experimentum crucis) by constructing a difference contrast.

Sounds interesting? Then take a look at some introductory literature, such as Henninger, Malejka, & Titz (2025), Furr (2004), Rosenthal & Rosnow (1985), Rosenthal, Rosnow, & Rubin (2000), or—for German-speaking readers---@sedlmeier2018. Contrast analysis is relatively easy to grasp if you are already familiar with ANOVA and correlation.

In this vignette, we assume a basic understanding of contrast analysis and focus on its practical application to a specific dataset. We first show how to install cofad and use its graphical user interface. We then work through several example analyses for between-subjects, within-subjects, and mixed designs in R.

Installation

Cofad has two components: the R package and a Shiny app that provides a graphical user interface.

If you only want to use the cofad app, no installation is required. A browser-only webR version is available at https://johannes-titz.github.io/cofad/. For the server-backed version, open cofad.titz.science in your browser; there you can select from categorized example datasets or upload your own data.

If you prefer the command-line interface or want to run the cofad app locally, install the package from CRAN:

install.packages("cofad")

Alternatively, you can install the development version from GitHub (you need the remotes package for this):

# install.packages("remotes") # uncomment if you do not have remotes installed
remotes::install_github("johannes-titz/cofad")

Now you can load cofad and use it directly in your R scripts.

You can also launch the app locally:

cofad::run_app()

Docker

To run the server-backed app in a container, build the included Dockerfile from the repository root and publish its port:

docker build -t cofad .
docker run --rm -p 3838:3838 cofad

Then open localhost:3838 in your browser. The image uses R 4.6.1, runs the app as a non-root user, and includes a container health check. To use another available Rocker Shiny version, override the build argument, for example docker build --build-arg R_VERSION=4.5.3 -t cofad ..

Using cofad

Before you start, your data must be in long format (also called narrow or tidy format). If you are unfamiliar with this format, see the short explanation in the Wikipedia article: https://en.wikipedia.org/wiki/Wide_and_narrow_data.

Graphical user interface

Load a .csv or .sav (SPSS) file, or select a packaged example grouped as between, within, or mixed. At startup, both the server-backed and browser-only webR interfaces load Rosenthal et al.’s Table 5.3 mixed-design example with its linear between- and within-subjects contrasts.

Load a data file

Select an example data set

cofad automatically examines replication and nesting to suggest the dependent variable, between-subjects factor, within-subjects factor, and participant ID. Every suggestion remains editable in the model table. The detector is deliberately conservative: it cannot infer the scientific meaning of variables or the intended hypothesis from correlations alone. Packaged examples therefore use the roles and planned weights documented in their books or papers, while still displaying the structural-detection result.

Automatically detected design

If detection is inconclusive or cannot run, the app does not force a structure. It retains a suggested dependent variable where possible, leaves uncertain roles as NONE, and asks you to choose the design manually in the same table. This fallback is available for between-, within-, and mixed-subjects designs.

Manual design selection after inconclusive detection

You can inspect the automatic suggestion from R with detect_design(your_data). It reports structural diagnostics and a confidence score, but its result should always be checked against the documented study design. In the app, manual role selection remains available regardless of the confidence score.

Between- and within-subjects contrast tables are activated automatically from the roles selected in the model table. For a mixed design, choose whether to test the Between × within contrast or the Within contrast averaged across groups. New mixed designs default to the former, as in Rosenthal et al.’s Table 5.3; packaged examples retain the analysis used by their source.

Select the contrast analysis

Specify the between-subjects contrast

Specify the within-subjects contrast

The default example comes from Rosenthal et al. (2000) (Table 5.3). The cognitive ability of nine children belonging to different age groups (between) was measured four times (within).

There are two hypotheses:

1. cognitive ability linearly increases over time (within) (\(\lambda_\mathrm{1} = -3, \lambda_\mathrm{2} = -1, \lambda_\mathrm{3} = 1, \lambda_\mathrm{4} = 3\))
2. cognitive ability linearly increases over age groups (between) (\(\lambda_\mathrm{Age 8} = -1, \lambda_\mathrm{Age 10} = 0, \lambda_\mathrm{Age12} = 1\))

Select the variables in the compact model table and edit the default contrast weights in the lambda tables. The app reports the contrast, a detailed variance decomposition with ordinary \(\eta^2\) for each SS component (including the descriptive within-group/error share) and partial eta squared for tested components, plus an interactive Plotly chart comparing total, between-group, and contrast-plus-error denominators. Its legend labels Contrast, Other between-group, and Within-group/error above the bars. Hovering over a segment shows its SS numerator and denominator. Small p values are shown numerically in scientific notation rather than as threshold statements. For mixed designs, this is explicitly a partition of participants’ derived within-contrast scores, not of the raw repeated outcome variance. The paper-ready report is copyable, while the F table can be exported as aligned plain text, rich HTML, or DOCX; numeric table columns are right-aligned. The Help and citation panel beneath the model contains the package references, README, and GitHub issue tracker.

Copy the paper-ready report

Inspect and export the F table

Explore the partition of total variation

A mixed design is ideal for trying the cofad app. You can construct a separate within-subjects model by setting the between-subjects factor to NONE. Likewise, set the within-subjects factor and participant ID to NONE to obtain a between-subjects model.

To compare two competing contrasts, select Compare two competing contrasts. Each active weight table then shows Favored and Rival columns. Cofad standardizes both vectors before analyzing Favored minus Rival, as lambda_diff() does. Initially, Rival reverses Favored, preserving the current test until you edit the rival hypothesis. Clearing the checkbox removes the Rival columns and immediately restores the Favored weights as ordinary single contrasts.

The graphical user interface will suffice for many users, but some will prefer to use the scripting capabilities of R. In the next sections we will look at several script examples for different designs.

Between-Subjects Designs

First, load the package:

library(cofad)

Now we need some data and hypotheses. We use the data from Furr (2004), which contain empathy ratings of students from different majors. The dataset is included in the cofad package:

data("furr_p4")
furr_p4
#>    empathy      major
#> 1       51 psychology
#> 2       56 psychology
#> 3       61 psychology
#> 4       58 psychology
#> 5       54 psychology
#> 6       62  education
#> 7       67  education
#> 8       57  education
#> 9       65  education
#> 10      59  education
#> 11      50   business
#> 12      49   business
#> 13      47   business
#> 14      45   business
#> 15      44   business
#> 16      50  chemistry
#> 17      45  chemistry
#> 18      40  chemistry
#> 19      49  chemistry
#> 20      41  chemistry

Furr proposes three hypotheses:

• Contrast A: Psychology majors have higher empathy scores than Education majors (\(\lambda_\mathrm{psych} = 1, \lambda_\mathrm{edu} = -1\)).
• Contrast B: Business majors have higher empathy scores than Chemistry majors (\(\lambda_\mathrm{bus} = 1, \lambda_\mathrm{chem} = -1\)).
• Contrast C: On average, Psychology and Education majors have higher empathy scores than Business and Chemistry majors (\(\lambda_\mathrm{psych} = 1, \lambda_\mathrm{edu} = 1, \lambda_\mathrm{bus} = -1, \lambda_\mathrm{chem} = -1\)).

These hypotheses involve simple mean comparisons, making them a good starting point. Let us use cofad to conduct the contrast analysis:

ca <- calc_contrast(dv = empathy, between = major,
                    lambda_between = c("psychology" = 1, "education" = -1,
                                       "business" = 0, "chemistry" = 0),
                    data = furr_p4)
ca
#>
#> We ran a contrast analysis for the following between contrasts: business = 0; chemistry = 0; education = -1; psychology = 1. This resulted in statistics of t(16) = -2.481; p = 0.9877 and an effect magnitude of rₑₛ = -0.276. Attention: Contrast fits in the opposite direction!

The print method provides basic information that can be used directly in a publication. The summary method provides additional details:

summary(ca)
#> Contrast Analysis Between
#>
#> Your contrast estimate is negative. This means that your data does not reflect the expected direction of your hypothesis specified by the contrast weights (lambdas).
#>
#> $Lambdas
#>   business  chemistry  education psychology
#>          0          0         -1          1
#>
#> $tTable
#>   L df      t p(t≥-2.481)①
#>  -6 16 -2.481        0.988
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $FTable
#>            SS df     MS     F      p
#> contrast   90  1 90.000 6.154 0.0246
#> within    234 16 14.625
#> total    1179 19
#>
#> $Effects
#>              effects
#> r_effectsize  -0.276
#> r_contrast    -0.527
#> r_alerting    -0.309

From this table, \(r_\mathrm{effectsize}\) is probably the most useful statistic. It is simply the correlation between the contrast weights (\(\lambda\)) and the dependent variable, which can also be calculated by hand:

lambdas <- rep(c(1, -1, 0, 0), each = 5)
cor(furr_p4$empathy, lambdas)
#> [1] -0.2762895

As you can see, the effect is negative, and cofad also warns that the observed contrast runs in the opposite direction to the hypothesis. This constitutes a clear failure of the hypothesis and points to substantial problems with the underlying theory.

The other two hypotheses can be tested in the same way.

ca <- calc_contrast(dv = empathy, between = major,
                    lambda_between = c("psychology" = 0, "education" = 0,
                                       "business" = 1, "chemistry" = -1),
                    data = furr_p4)
ca
#>
#> We ran a contrast analysis for the following between contrasts: business = 1; chemistry = -1; education = 0; psychology = 0. This resulted in statistics of t(16) = 0.827; p = 0.2102 and an effect magnitude of rₑₛ = 0.092.
ca <- calc_contrast(dv = empathy, between = major,
                    lambda_between = c("psychology" = 1, "education" = 1,
                                       "business" = -1, "chemistry" = -1),
                    data = furr_p4)
ca
#>
#> We ran a contrast analysis for the following between contrasts: business = -1; chemistry = -1; education = 1; psychology = 1. This resulted in statistics of t(16) = 7.601; p = 5.349e-07 and an effect magnitude of rₑₛ = 0.847.

When you compare these results with those reported in Furr (2004), you will find the same pattern. Note that a one-sided test is appropriate because the direction of the contrast is specified in advance. A contrast hypothesis should not be treated as interchangeable with its reverse direction. The F test itself is non-directional; for a single contrast, the corresponding directional test uses the signed t statistic, with \(|t| = \sqrt{F}\).

Now, imagine a more interesting hypothesis that goes beyond simple mean differences. Suppose an elaborate theory predicts means of 73, 61, 51, and 38 for the four groups. We can test this pattern directly with cofad, because the package automatically centers the contrast weights so that their mean is zero:

ca <- calc_contrast(dv = empathy, between = major,
                    lambda_between = c("psychology" = 73, "education" = 61,
                                       "business" = 51, "chemistry" = 38),
                    data = furr_p4)
#> lambdas are centered and rounded to 3 digits
ca
#>
#> We ran a contrast analysis for the following between contrasts: business = -4.75; chemistry = -17.75; education = 5.25; psychology = 17.25. This resulted in statistics of t(16) = 6.121; p = 7.375e-06 and an effect magnitude of rₑₛ = 0.682.

The manual calculation yields the same effect size:

lambdas <- rep(c(73, 61, 51, 38), each = 5)
cor(furr_p4$empathy, lambdas)
#> [1] 0.6817294

Let us now turn to contrast analysis for within-subjects designs.

Within-Subjects Designs

For within-subjects designs, the calculations differ substantially, but cofad takes care of the details. We only need to use the within-subject parameters within and lambda_within instead of their between-subjects counterparts. As an example, we use Table 16.5 from Sedlmeier & Renkewitz (2018), in which reading ability was assessed for eight participants under four different conditions. The hypothesis is that reading performance is best without music, lower under white noise, and lowest with music, regardless of the type of music.

data("sedlmeier_p537")
head(sedlmeier_p537)
#>   reading_test participant         music
#> 1           27           1 without music
#> 2           25           2 without music
#> 3           30           3 without music
#> 4           29           4 without music
#> 5           30           5 without music
#> 6           33           6 without music
within <- calc_contrast(dv = reading_test, within = music,
                        lambda_within = c("without music" = 1.25,
                                          "white noise" = 0.25,
                                          "classic" = -0.75,
                                          "jazz" = -0.75),
                        id = participant, data = sedlmeier_p537)
summary(within)
#> Contrast Analysis Within
#>
#> $Lambdas
#>       classic          jazz   white noise without music
#>         -0.75         -0.75          0.25          1.25
#>
#> $tTable
#>  mean of L    SE df     t p(t≥5.269)① 95%CI-lower 95%CI-upper
#>      5.875 1.115  7 5.269    0.000581       3.238       8.512
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $Effects
#>
#> r-contrast 0.894
#> g-contrast 1.863
within
#>
#> We ran a within-subjects contrast analysis using participant-level L scores (weighted sums retaining response magnitude) for the following contrast weights: classic = -0.75; jazz = -0.75; white noise = 0.25; without music = 1.25. This resulted in t(7) = 5.269; p = 5.810e-04 and an effect magnitude of r_contrast = 0.894.

The significance test is one-tailed, and the paper-ready report provides the signed effect-size correlation \(r_\mathrm{contrast}\).

For this within-subjects contrast, \(r_\mathrm{contrast}\) is the signed correlation-form effect derived from the same test, \(r_\mathrm{contrast} = \operatorname{sign}(t)\sqrt{t^2/(t^2 + df_\mathrm{error})}\). Consequently, \(r_\mathrm{contrast}^2\) equals the contrast-specific \(\eta^2\) and partial \(\eta_p^2\). The between-design \(r_\mathrm{es}\) and \(r_\mathrm{alerting}\) are not defined by this pure within-subjects contrast partition.

We can verify the result by hand relatively easily:

mtr <- matrix(sedlmeier_p537$reading_test, ncol = 4)
lambdas <- c(1.25, 0.25, -0.75, -0.75)
lc1 <- mtr %*% lambdas
t.test(lc1, alternative = "greater")
#>
#>  One Sample t-test
#>
#> data:  lc1
#> t = 5.2689, df = 7, p-value = 0.000581
#> alternative hypothesis: true mean is greater than 0
#> 95 percent confidence interval:
#>  3.762478      Inf
#> sample estimates:
#> mean of x
#>     5.875

All that is needed is the linear combination of the dependent variable and the contrast weights for each participant. These values are then tested against zero using a standard \(t\)-test. Although this can be done manually, cofad is faster and provides additional information.

Using participant L or r scores

For a within contrast, within_score = "L" (the default) calculates each participant’s weighted sum. It retains the absolute magnitude of the response, so choose L when the size of a predicted increase or decrease is substantively meaningful. within_score = "r" instead correlates each participant’s response profile with the contrast weights. It measures pattern agreement and is unchanged by adding a constant or multiplying a profile by a positive constant. Choose r when adherence to the predicted shape matters more than amplitude, especially when there are many repeated conditions.

The choice should follow the research question and be made before comparing the results. With only two repeated levels, a defined r score can only be -1 or 1 and discards magnitude; L is usually more informative. An r score is undefined for a participant whose responses are all equal, and cofad reports those IDs.

Rosenthal et al.’s Table 5.4 makes the distinction concrete. Child A has the larger weighted sum but Child B follows the linear pattern perfectly:

data("rosenthal_tbl54")
weights <- c("1" = -3, "2" = -1, "3" = 1, "4" = 3)

calc_contrast(dv, within = within, lambda_within = weights, id = id,
              data = rosenthal_tbl54, within_score = "L")$participant_scores
#>  A  B
#> 18 10
calc_contrast(dv, within = within, lambda_within = weights, id = id,
              data = rosenthal_tbl54, within_score = "r")$participant_scores
#>       A       B
#> 0.56921 1.00000

For a complete within-subjects analysis based on participant correlations, change only within_score:

within_r <- calc_contrast(
  dv = reading_test,
  within = music,
  lambda_within = c(
    "without music" = 1.25,
    "white noise" = 0.25,
    "classic" = -0.75,
    "jazz" = -0.75
  ),
  id = participant,
  data = sedlmeier_p537,
  within_score = "r"
)
summary(within_r)
#> Contrast Analysis Within
#>
#> $Lambdas
#>       classic          jazz   white noise without music
#>         -0.75         -0.75          0.25          1.25
#>
#> $tTable
#>  mean of r    SE df      t p(t≥12.544)① 95%CI-lower 95%CI-upper
#>       0.81 0.065  7 12.544     2.36e-06       0.658       0.963
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $Effects
#>
#> r-contrast 0.978
#> g-contrast 4.435

The app exposes the same choice as Participant-level within score, with labels that distinguish magnitude (L) from pattern fit (r).

Mixed Designs

A mixed design combines between- and within-subjects factors. Cofad first calculates one participant-level score for the within contrast: a weighted sum (L) by default or a profile correlation (r) when within_score = "r". That score becomes the dependent variable for the between-subjects contrast.

We will again use the example from Rosenthal et al. (2000) (see the graphical-user-interface section). The cognitive ability of nine children from three different age groups (between-subjects factor) was measured at four time points (within-subjects factor).

There are two hypotheses:

1. Cognitive ability increases linearly over time (within subjects): (\(\lambda_\mathrm{1} = -3, \lambda_\mathrm{2} = -1, \lambda_\mathrm{3} = 1, \lambda_\mathrm{4} = 3\)).
2. Cognitive ability increases linearly across age groups (between subjects): (\(\lambda_\mathrm{Age 8} = -1, \lambda_\mathrm{Age 10} = 0, \lambda_\mathrm{Age 12} = 1\)).

Let us have a look at the data and the corresponding calculations:

data("rosenthal_tbl53")
head(rosenthal_tbl53)
#>   dv between id within
#> 1  3    age8  1      1
#> 2  1    age8  2      1
#> 3  4    age8  3      1
#> 4  4   age10  4      1
#> 5  5   age10  5      1
#> 6  5   age10  6      1
lambda_within <- c("1" = -3, "2" = -1, "3" = 1, "4" = 3)
lambda_between <-c("age8" = -1, "age10" = 0, "age12" = 1)

contr_mx <- calc_contrast(dv = dv,
                          between = between,
                          lambda_between = lambda_between,
                          within = within,
                          lambda_within = lambda_within,
                          id = id,
                          data = rosenthal_tbl53)
contr_mx
#>
#> We ran a mixed contrast analysis using participant-level L scores (weighted sums retaining response magnitude) for the within-subjects contrast and the following between-subjects contrast weights: age10 = 0; age12 = 1; age8 = -1. This resulted in t(6) = 4.496; p = 0.002062 and an effect magnitude of rₑₛ = 0.871.

The results resemble those from a between-subjects contrast analysis. The summary additionally reports the effect sizes, along with the between-group means and standard errors of the selected participant-level score.

summary(contr_mx)
#> Contrast Analysis Mixed
#>
#> $Lambdas
#> age10 age12  age8
#>     0     1    -1
#>
#> $tTable
#>      L df     t p(t≥4.496)①
#>  5.333  6 4.496     0.00206
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $FTable
#>              SS df     MS      F       p
#> contrast 42.667  1 42.667 20.211 0.00412
#> within   12.667  6  2.111
#> total    56.222  8
#>
#> $Effects
#>              effects
#> r_effectsize   0.871
#> r_contrast     0.878
#> r_alerting     0.990

To test pattern fit rather than response magnitude, use the same model with within_score = "r". The returned participant_scores retain the unrounded correlations. Rosenthal et al. printed these values to two decimals before their worked calculation, so calculations from the raw data differ slightly from the book’s rounded F value.

contr_mx_r <- calc_contrast(
  dv = dv, between = between, lambda_between = lambda_between,
  within = within, lambda_within = lambda_within, id = id,
  data = rosenthal_tbl53, within_score = "r"
)
contr_mx_r$participant_scores
#>         1         2         3         4         5         6         7         8
#> 0.2581989 0.4472136 0.7745967 0.6741999 0.4472136 0.6741999 0.9438798 0.9233805
#>         9
#> 0.9486833

Comparing two hypotheses

With cofad, you can also compare two competing hypotheses directly. A useful introduction to this logic is provided by Henninger et al. (2025). As an example, Sedlmeier & Renkewitz (2013) use a fictitious data set on the problem-solving skills of boys:

sedlmeier_p525
#>    lsg between lambda
#> 1    1      KT     -2
#> 2    2      KT     -2
#> 3    2      KT     -2
#> 4    2      KT     -2
#> 5    3      KT     -2
#> 6    4      JT      3
#> 7    2      JT      3
#> 8    3      JT      3
#> 9    4      JT      3
#> 10   3      JT      3
#> 11   2      MT     -1
#> 12   3      MT     -1
#> 13   3      MT     -1
#> 14   1      MT     -1
#> 15   2      MT     -1

Here, lsg is the number of solved exercises, and the groups are KT = no training, JT = boys-specific training, and MT = girls-specific training. Two hypotheses are competing:

• -2, 3, -1 (boys benefit from boys-specific training)
• -2, 1, 1 (boys benefit from training, independently of the type of training)

First, we need to create the difference lambdas:

lambda1 <- c(-2, 3, -1)
lambda2 <- c(-2, 1, 1)
lambda <- lambda_diff(lambda1, lambda2, labels = c("KT", "JT", "MT"))
lambda
#>         JT         KT         MT
#>  0.6816234  0.4883935 -1.1700168

Note that you cannot simply subtract the contrast weights, because the two sets of weights may differ in variance, which affects the comparison. Instead, the weights must first be standardized. lambda_diff takes care of this automatically.

Now you can run a standard contrast analysis:

ca_competing <- calc_contrast(
  dv = lsg,
  between = between,
  lambda_between = round(lambda, 2),
  data = sedlmeier_p525
)
#> lambdas are centered and rounded to 3 digits
summary(ca_competing)
#> Contrast Analysis Between
#>
#> $Lambdas
#>    JT    KT    MT
#>  0.68  0.49 -1.17
#>
#> $tTable
#>      L df     t p(t≥1.136)①
#>  0.582 12 1.136       0.139
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $FTable
#>              SS df    MS     F     p
#> contrast  0.818  1 0.818 1.291 0.278
#> within    7.600 12 0.633
#> total    11.733 14
#>
#> $Effects
#>              effects
#> r_effectsize   0.264
#> r_contrast     0.312
#> r_alerting     0.445
ca_competing
#>
#> We ran a contrast analysis for the following between contrasts: JT = 0.68; KT = 0.49; MT = -1.17. This resulted in statistics of t(12) = 1.136; p = 0.139 and an effect magnitude of rₑₛ = 0.264.

Here, we rounded the contrast weights so that the results closely match those reported in Sedlmeier & Renkewitz (2013), who found \(t = 1.137\) and \(r_\mathrm{effectsize} = 0.26\). The effect size is identical. For the \(t\) statistic, we take the signed square root of the \(F\) value, 1.291, which gives 1.136. The remaining difference from the reported value of 1.137 is due to rounding.

The same approach also works for within-subjects designs. We can use the reading-comprehension data from above as an example, in which reading ability was assessed for eight participants under four different conditions:

sedlmeier_p537
#>    reading_test participant         music
#> 1            27           1 without music
#> 2            25           2 without music
#> 3            30           3 without music
#> 4            29           4 without music
#> 5            30           5 without music
#> 6            33           6 without music
#> 7            31           7 without music
#> 8            35           8 without music
#> 9            25           1   white noise
#> 10           26           2   white noise
#> 11           32           3   white noise
#> 12           29           4   white noise
#> 13           28           5   white noise
#> 14           30           6   white noise
#> 15           32           7   white noise
#> 16           34           8   white noise
#> 17           21           1       classic
#> 18           25           2       classic
#> 19           23           3       classic
#> 20           26           4       classic
#> 21           27           5       classic
#> 22           26           6       classic
#> 23           29           7       classic
#> 24           31           8       classic
#> 25           23           1          jazz
#> 26           24           2          jazz
#> 27           24           3          jazz
#> 28           28           4          jazz
#> 29           24           5          jazz
#> 30           26           6          jazz
#> 31           27           7          jazz
#> 32           32           8          jazz

There are two competing hypotheses:

• \(1.25, 0.25, -0.75, -0.75\): Reading performance is highest without music, lower with white noise, and lowest with music, regardless of music type.
• \(3, -1, -1, -1\): Any kind of noise reduces reading performance to the same extent.

Again, we first need to calculate the difference contrast weights:

lambda1 <- c(1.25, 0.25, -0.75, -0.75)
lambda2 <- c(3, -1, -1, -1)
lambda <- lambda_diff(lambda2, lambda1,
                      labels = c("without music", "white noise", "classic",
                                 "jazz"))
lambda
#>       classic          jazz   white noise without music
#>     0.3271838     0.3271838    -0.8788616     0.2244941

Note that we use lambda2 as the first argument to lambda_diff, because this follows the calculation in Sedlmeier & Renkewitz (2013), where the difference is defined as hypothesis 2 minus hypothesis 1.

Now we can run the contrast analysis:

contr_wi <- calc_contrast(
  dv = reading_test,
  within = music,
  lambda_within = round(lambda, 2),
  id = participant,
  data = sedlmeier_p537
)
#> lambdas are centered and rounded to 3 digits
summary(contr_wi)
#> Contrast Analysis Within
#>
#> Your contrast estimate is negative. This means that your data does not reflect the expected direction of your hypothesis specified by the contrast weights (lambdas).
#>
#> $Lambdas
#>       classic          jazz   white noise without music
#>          0.33          0.33         -0.88          0.22
#>
#> $tTable
#>  mean of L    SE df     t p(t≥-3.77)① 95%CI-lower 95%CI-upper
#>       -2.2 0.584  7 -3.77       0.997       -3.58       -0.82
#>
#> ①The p-value refers to a one-tailed test.
#>
#> $Effects
#>
#> r-contrast -0.819
#> g-contrast -1.333
contr_wi
#>
#> We ran a within-subjects contrast analysis using participant-level L scores (weighted sums retaining response magnitude) for the following contrast weights: classic = 0.33; jazz = 0.33; white noise = -0.88; without music = 0.22. This resulted in t(7) = -3.77; p = 0.9965 and an effect magnitude of r_contrast = -0.819. Attention: Contrast fits in the opposite direction!

Sedlmeier & Renkewitz (2013) reported a \(t\) value of \(-3.75\) and a \(g_\mathrm{contrast}\) of \(-1.33\). Again, the slight difference in the \(t\) value compared with our calculation is due to rounding. Because the difference contrast was defined as hypothesis 2 minus hypothesis 1, the negative test statistic and effect size indicate that hypothesis 1 provides the better fit.

Aggregated Data

Sometimes, you may want to run a contrast analysis on aggregated data, for example when the raw data are unavailable. If you have the means, standard deviations, and sample sizes for each condition, you can still perform the analysis with cofad. For instance, if we aggregate the data from our first example, we can reproduce the contrast analysis from the summary statistics alone:

furr_agg <- aggregate(
  empathy ~ major, furr_p4,
  function(x) c(mean = mean(x), sd = sd(x), n = length(x))
)
furr_agg <- data.frame(
  major = furr_agg$major,
  mean = furr_agg$empathy[, "mean"],
  sd = furr_agg$empathy[, "sd"],
  n = furr_agg$empathy[, "n"]
)
lambdas = c("psychology" = 1, "education" = -1, "business" = 0, "chemistry" = 0)
calc_contrast_aggregated(mean, sd, n, major, lambdas, furr_agg)
#>
#> We ran a contrast analysis for the following between contrasts: business = 0; chemistry = 0; education = -1; psychology = 1. This resulted in statistics of t(16) = -2.481; p = 0.9877 and an effect magnitude of rₑₛ = -0.276. Attention: Contrast fits in the opposite direction!

And indeed, the result is identical to that obtained from the raw data:

ca <- calc_contrast(dv = empathy, between = major,
                    lambda_between = c("psychology" = 1, "education" = -1,
                                       "business" = 0, "chemistry" = 0),
                    data = furr_p4)
ca
#>
#> We ran a contrast analysis for the following between contrasts: business = 0; chemistry = 0; education = -1; psychology = 1. This resulted in statistics of t(16) = -2.481; p = 0.9877 and an effect magnitude of rₑₛ = -0.276. Attention: Contrast fits in the opposite direction!

Note that this approach is limited to between-subjects designs.

Testing

The test suite includes unit, regression, example, summary-method, validation, in-process Shiny server, and browser integration tests. GitHub Actions runs both R CMD check and an in-process coverage workflow on every push and pull request. The server tests use shiny::testServer(), so covr::package_coverage(type = "tests") measures app logic without starting a browser process. Browser tests remain a thinner end-to-end layer and are skipped on CRAN and continuous integration because their timing has been unreliable on some runners.

Browser-only webR version

A static Shinylive build can run cofad entirely in the browser through webR, without a Shiny server. It is published to https://johannes-titz.github.io/cofad/ by the deploy-shinylive GitHub Actions workflow on every push to the main branch.

For a local build, run source("tools/build-shinylive.R") from the repository root; the generated docs/ directory can be served by any static web host. The current static artifact is approximately 65 MiB. In a local Chromium benchmark on August 15, 2026, a fresh isolated browser context showed the controls after 26.9 seconds and a selected example’s model table after 32.9 seconds. A cached reload took 22.4 and 26.8 seconds, respectively. These are single-machine localhost measurements, not network guarantees. The server-backed app remains preferable when the lowest latency matters.

Issues and Support

If you find any bugs, please use the issue tracker at:

https://github.com/johannes-titz/cofad/issues

If you need answers on how to use the package, drop an e-mail at johannes at titz.science or johannes.titz at gmail.com

Contributing

Comments and feedback of any kind are very welcome! We carefully consider every suggestion for improving the code, documentation, and examples. Even small contributions, such as clearer wording or grammar corrections, are greatly appreciated.

If you would like to submit a pull request, please make sure that the package still builds without errors, warnings, or notes. In general, follow the recommendations in the R Packages book: https://r-pkgs.org/ and the tidyverse style guide: https://style.tidyverse.org/.

Acknowledgments

We would like to thank Thomas Schäfer and Isabell Winkler for testing cofad and providing valuable feedback.

References