A taichi grid is a superposition comparison: the two sources share one
position, which makes "are these similar?" and "which is bigger here?" easy
to see and "by how much?" impossible. taichi_summary() is the tidy answer
to the last question — the numbers behind the glyph, one row per input
row, for the reader who needs a table rather than a picture.
Arguments
- data
A data frame, the same one passed to
ggplot2::ggplot().- yin, yang
Unquoted column names (or strings naming columns) for the two sources, exactly as in
geom_taichi().- x, y
Optional unquoted column names identifying each cell. When given they are carried through to the output as the first columns, which makes the result joinable back onto the plotting data.
Value
A data frame with one row per row of data: the optional x and
y cell identifiers, yin, yang, difference, ratio, log_ratio,
z, dominant and rank.
The statistics
differenceyin - yang. Positive means the yin fish (the top bulb) carries the larger value.ratioyin / yang, andNAwherever either value is not strictly positive — a ratio of a negative or zero quantity is not a ratio, andInfis never returned.log_ratiolog2(yin / yang), so a value of 1 means yin is twice yang and -1 means half. Symmetric around zero, whichratiois not, and the right choice when the two sources span orders of magnitude.zThe difference of the two standardised sources: each column is centred and scaled across the whole grid, then subtracted. This is the statistic to use when the two sources are not in the same units, because it asks "which source is unusually high here, relative to its own spread?" rather than "which number is bigger?".
dominantWhich source is larger in that cell, named after the columns supplied, or
"tie".rankThe cell's rank by size of
abs(difference), 1 being the widest gap. Cells with a missing difference getNA.
A caveat on rank
A grid of 96 cells is 96 implicit comparisons, and the most extreme cell in
a grid that size is frequently the most extreme noise rather than the
largest real effect. Treat rank as a list of places to look, not as a
list of findings: taking the top-ranked cell and reporting it is the
multiple-comparisons problem in miniature. With one observation per cell
there is no per-cell test to correct with, so the honest check is on the
field as a whole.
See also
geom_taichi()'s explicit argument, which shows one of these
statistics inside the glyph, and geom_taichi_diff(), which draws it as
a diverging heatmap.
Examples
summ <- taichi_summary(cafes_tg, yin = matcha, yang = espresso,
x = week, y = neighbourhood)
head(summ)
#> x y yin yang difference ratio log_ratio z dominant rank
#> 1 1 Old Town 32.8 48.5 -15.7 0.6762887 -0.5642889 -1.3891789 espresso 55
#> 2 2 Old Town 40.5 52.9 -12.4 0.7655955 -0.3853458 -1.1336049 espresso 66
#> 3 3 Old Town 38.4 50.9 -12.5 0.7544204 -0.4065593 -1.1321353 espresso 63
#> 4 4 Old Town 36.6 45.2 -8.6 0.8097345 -0.3044791 -0.7762886 espresso 72
#> 5 5 Old Town 38.3 56.2 -17.9 0.6814947 -0.5532257 -1.6120500 espresso 48
#> 6 6 Old Town 41.1 60.0 -18.9 0.6850000 -0.5458241 -1.7148359 espresso 45
# the five widest gaps -- places to look, not findings; see the caveat above
head(summ[order(summ$rank), ], 5)
#> x y yin yang difference ratio log_ratio z dominant
#> 35 11 University 74.1 24.0 50.1 3.087500 1.626439 4.260819 matcha
#> 82 10 Garden District 70.9 22.2 48.7 3.193694 1.675226 4.152065 matcha
#> 33 9 University 76.1 28.4 47.7 2.679577 1.422006 4.037432 matcha
#> 84 12 Garden District 72.2 24.7 47.5 2.923077 1.547488 4.038890 matcha
#> 36 12 University 76.1 32.9 43.2 2.313070 1.209809 3.637092 matcha
#> rank
#> 35 1
#> 82 2
#> 33 3
#> 84 4
#> 36 5