move new example to datasets

contrasts_kwdyz11.qmd

@@ -10,18 +10,17 @@ fig-format: png
 
 ```{julia}
 #| code-fold: true
+using AlgebraOfGraphics
 using CairoMakie
 using Chain
 using DataFrames
 using MixedModels
-using RCall
 using SMLP2024: dataset
 using StatsBase
 using StatsModels
 
 CairoMakie.activate!(; type="png")
 
-import ProgressMeter
 progress = false
 ```
 
@@ -30,13 +29,13 @@ Many researchers have pointed out that contrasts should be "tested instead of, r
 
 For a (quasi-)experimental set of data, there is (or should be) a clear _a priori_ theoretical commitment to specific hypotheses about differences between factor levels and how these differences enter in interactions with other factors. This specification should be used in the first LMM and reported, irrespective of the outcome. If alternative theories lead to alternative _a priori_ contrast specifications, both analyses are justified. If the observed means render the specification completely irrelevant, the comparisons originally planned could still be reported in a Supplement).
 
-In this script, we are working through a large number of different contrasts for the same data. The purpose is to introduce both the preprogrammed (“canned”) and the general options to specify hypotheses about main effects and interactions. Obviously, we do not endorse generating a plot of the means and specifying the contrasts accordingly. This is known as the [Texas sharpshooter](https://www.bayesianspectacles.org/origin-of-the-texas-sharpshooter/) fallacy. The link leads to an illustration and brief historical account by Wagenmakers (2018).
+In this script, we are working through a large number of different contrasts for the same data. The purpose is to introduce both the preprogrammed ("canned") and the general options to specify hypotheses about main effects and interactions. Obviously, we do not endorse generating a plot of the means and specifying the contrasts accordingly. This is known as the [Texas sharpshooter](https://www.bayesianspectacles.org/origin-of-the-texas-sharpshooter/) fallacy. The link leads to an illustration and brief historical account by Wagenmakers (2018).
 
 Irrespective of how results turn out, there is nothing wrong with specifying a set of post-hoc contrasts to gain a better understanding of what the data are trying to tell us. Of course, in an article or report about the study, the _a priori_ and post-hoc nature of contrast specifications must be made clear. Some kind of alpha-level adjustment (e.g., Bonferroni) may be called for, too. And, of course, there are grey zones.
 
-There is quite a bit of statistical literature on contrasts. Two “local” references are @Brehm2022 and @Schad2020.
+There is quite a bit of statistical literature on contrasts. Two "local" references are @Brehm2022 and @Schad2020.
 
-For further readings see “Further Readings” in @Schad2020.
+For further readings see "Further Readings" in @Schad2020.
 
 # Example data {#sec-data}
 
@@ -439,10 +438,7 @@ Three factors:
 2 x 3 = 6 measures / subject
 
 ```{julia}
-R"""
-dat2 = readRDS("data/Exp_2x2x3.rds");
-"""
-@rget(dat2)
+dat2 = dataset(:exp_2x2x3)
 ```
 
 We select an LMM supported by the data. 
@@ -482,35 +478,20 @@ A: A2 & B: B3               0.523243      0.950202   0.55    0.5819
 ───────────────────────────────────────────────────────────────────
 ```
 
-The  following figure also appears only in interactive chunk execution. The chunk generates also an error when rendered.
-
-```{r}
-#| eval: true
-
-library(tidyverse)
-dat2 = readRDS("data/Exp_2x2x3.rds");
-
-cbPalette <- c("#000000", "#E69F00", "#56B4E9", "#009E73",
-               "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
-
-tbl1 <- 
-  dat2 |> 
-  group_by(Subj, A, B) |> 
-  reframe(N=n(),dv=mean(dv)) |> 
-  group_by(A, B) |> 
-  reframe(N=n(), dv_M=mean(dv), dv_SD=sd(dv), dv_SE=dv_SD/sqrt(N))
-tbl1
-
-fig1 <- 
-  tbl1 |> 
-  ggplot(aes(y=dv_M, x=B, group=A, color=A)) + 
-  geom_point() + 
-  geom_line() +
-  scale_color_manual(values=cbPalette[2:3]) +
-  theme_bw()
-
-print(fig1)
-NULL
+```{julia}
+using Chain
+tbl1 = @chain DataFrame(dat2) begin
+  groupby(_, [:Subj, :A, :B])
+  combine(_, nrow => :n, :dv => mean => :dv)
+  groupby(_, [:A, :B])
+  combine(_, 
+          :dv => mean => :dv_M,
+          :dv => std => :dv_SD,
+          :dv => sem => :dv_SE)
+end 
+
+fig1 = data(tbl1) * mapping(:B, :dv_M; color=:A) * (visual(Lines) + visual(Scatter))
+draw(fig1)
 ```
 
 

src/datasets.jl

@@ -2,7 +2,7 @@ _file(x) = joinpath(CACHE[], string(x, ".arrow"))
 
 clear_scratchspaces!() = Scratch.clear_scratchspaces!(@__MODULE__)
 
-const datasets = 
+const DATASETS = 
     CSV.read(
         IOBuffer(
 """
@@ -21,6 +21,7 @@ fggk21_Child,c2fmn,1,61c91e00336e6f804e9f6b86986ebb4a14561cc4908b3a21cb27c113d2b
 fggk21_Score,7fqx3,1,99d73ee705aaf5f4ee696eadbba992d0113ba6f467ce337a62a63853e4617400
 kkl15,p8cea,2,90d7bb137c8613d7a15c8597c461aee7c7cb0f0989a07c80fc93e1fbe2e5c156
 kwdyz11,4cv52,3,2fa23aa8aa25e1adb10183c8d29646ae0d19d6baef9d711c9906f7fa1b225571
+exp_2x2x3,za9gs,1,cb09684b7373492e849c83f20a071b97f986123677134ac2ddb9ec0dcb32e503
 """
         ),
         Table;
@@ -29,25 +30,34 @@ kwdyz11,4cv52,3,2fa23aa8aa25e1adb10183c8d29646ae0d19d6baef9d711c9906f7fa1b225571
     )
 
 if @isdefined(_cacheddatasets)
-    empty!(_cacheddatasets)    # start from an empty cache in case datasets has changed
+    empty!(_cacheddatasets)    # start from an empty cache in case DATASETS has changed
 else
     const _cacheddatasets = Dict{Symbol, Arrow.Table}()
 end
 
+"""
+    datasets()
+
+Return a vector of the names of datasets available for use in [`dataset`](@ref).    
+"""
+function datasets()
+    return sort!(vcat(SMLP2024.DATASETS.dsname, MixedModelsDatasets.datasets()))
+end
+
 """
     dataset(name::Union(Symbol, AbstractString))
 
 Return as an `Arrow.Table` the dataset named `name`.
 
 Available dataset names, their versions, the filenames on the osf.io site and an SHA2 checksum of their contents
-are in the table `datasets`.
+are in the table `DATASETS`.
 
 The files are cached in the scratchspace for this package.  The name of this directory is the value of `CACHE[]`.
 """
 function dataset(nm::AbstractString)
     return get!(_cacheddatasets, Symbol(nm)) do  # retrieve from cache if available, otherwise
-        # check for nm in datasets table first so MMDS can be overridden
-        rows = filter(==(nm) ∘ getproperty(:dsname), datasets)
+        # check for nm in DATASETS table first so MMDS can be overridden
+        rows = filter(==(nm) ∘ getproperty(:dsname), DATASETS)
         if isempty(rows)
             nm in MMDS || error("Dataset '$nm' is not available")
             MixedModelsDatasets.dataset(nm)
@@ -58,10 +68,12 @@ function dataset(nm::AbstractString)
                 if ismissing(row.filename)
                     load_quiver()  # special-case `ratings` and `movies`
                 else
+                    @info "Downloading dataset..."
                     Downloads.download(
                         string("https://osf.io/", row.filename, "/download?version=", row.version),
                         fnm,
                     )
+                    @info "done"
                 end
             end
             if row.sha2 ≠ bytes2hex(open(sha2_256, fnm))