27 Displaying Data: Tables

If a graph is for seeing a pattern, a table is for reading an exact value. Ask which one your reader needs. If they need to feel the shape of a trend, draw it. If they need to look up “what was the bust rate for the Jets, exactly?” or compare a handful of precise numbers side by side, give them a table. A table is not a failed graph; it is a different instrument for a different job, and, like a graph, it can be honest or it can bury the point under too many digits.

We continue with the same real data from The Sports Page we used for graphics, so you can feel the difference between the two instruments on identical numbers.

27.1 Learning Objectives

Decide when a table serves the reader better than a graph.
Round to meaningful precision - kill false precision.
Order rows to carry information, and align numbers so they can be compared.
Summarize to the question, rather than dumping the raw data.

27.2 When a Table Beats a Graph

A table wins when the numbers are few, the exact values matter, or the reader will look things up rather than scan for a shape. The strikeout trend was a job for a line. But “what did each decade actually look like, in numbers I can quote?” is a job for a small table. Summarize the 42 years down to five decade averages - few rows, exact values, quotable:

mlb <- read.csv("data/mlb_league_by_year.csv")
mlb$decade <- paste0(floor(mlb$year / 10) * 10, "s")

dec <- aggregate(cbind(K_9, BB_9, HR_9, AVG) ~ decade, data = mlb, FUN = mean)
knitr::kable(dec, digits = c(0, 2, 2, 2, 3),
             col.names = c("Decade", "K/9", "BB/9", "HR/9", "AVG"),
             caption = "MLB by the decade: the strikeout climb, in exact numbers.")

MLB by the decade: the strikeout climb, in exact numbers.
Decade	K/9	BB/9	HR/9	AVG
1980s	5.70	3.30	0.87	0.257
1990s	6.18	3.45	0.96	0.264
2000s	6.62	3.38	1.08	0.265
2010s	7.87	3.12	1.08	0.254
2020s	8.69	3.31	1.19	0.244

Five rows, and you can quote any cell. The graph made you feel the strikeout climb; the table lets you say “K/9 went from about 5.7 in the 1980s to about 8.5 in the 2020s” without guessing off an axis. Same data, different instrument, different job.

27.3 Kill False Precision

The fastest way to make a table mislead, or simply to tire your reader, is to print every digit the computer produced. A bust rate stored as 0.5455 does not mean you know it to four decimals; with a few dozen draft picks, you barely know the second. Rounding is not sloppiness; it is honesty about how much you actually know. Watch the difference:

nfl <- read.csv("data/nfl_draft_by_team.csv")
worst <- nfl[order(-nfl$rate_bust_a), c("team", "n_picks", "rate_bust_a", "rate_clean_hit")]

# false precision: every digit the machine produced
knitr::kable(head(worst, 4), row.names = FALSE,
             caption = "Too many digits — implies a precision the data don't have.")

Too many digits — implies a precision the data don’t have.
team	n_picks	rate_bust_a	rate_clean_hit
LV	44	0.7273	0.2500
CIN	52	0.7174	0.2609
GB	49	0.6905	0.2381
NYG	49	0.6829	0.3171

# honest precision: percentages, rounded to what the sample can support
worst$rate_bust_a   <- round(100 * worst$rate_bust_a)
worst$rate_clean_hit <- round(100 * worst$rate_clean_hit)
knitr::kable(head(worst, 4), row.names = FALSE,
             col.names = c("Team", "Picks", "Bust rate (%)", "Clean-hit rate (%)"),
             caption = "Rounded to whole percentages — easier to read, honest about precision.")

Rounded to whole percentages — easier to read, honest about precision.
Team	Picks	Bust rate (%)	Clean-hit rate (%)
LV	44	73	25
CIN	52	72	26
GB	49	69	24
NYG	49	68	32

The second table is easier to read and more truthful. Show the digits you can defend, and not one more.

27.4 Order Rows to Carry Information

A table’s row order is free information - spend it. An alphabetical list of teams makes the reader hunt; a list sorted by the thing you care about answers the question before they ask it. We already sorted by bust rate above (worst first), so the ranking is the message. Now show the two ends that a reader actually wants - the best and worst drafters - instead of dumping all 32:

ends <- rbind(
  cbind(group = "Worst drafters", head(worst[, c("team", "n_picks", "rate_bust_a")], 5)),
  cbind(group = "Best drafters",  tail(worst[, c("team", "n_picks", "rate_bust_a")], 5))
)
knitr::kable(ends, row.names = FALSE,
             col.names = c("", "Team", "Picks", "Bust rate (%)"),
             caption = "The two ends of the ranking — the comparison the reader wanted.")

The two ends of the ranking — the comparison the reader wanted.
	Team	Picks	Bust rate (%)
Worst drafters	LV	44	73
Worst drafters	CIN	52	72
Worst drafters	GB	49	69
Worst drafters	NYG	49	68
Worst drafters	LAR	51	68
Best drafters	BUF	46	48
Best drafters	MIN	48	48
Best drafters	KC	42	47
Best drafters	BAL	33	40
Best drafters	WAS	35	38

Ten rows, sorted, grouped into the contrast that matters. A 32-row alphabetical dump would contain the same facts but communicate far less of them. Order and trim the table to the question; do not make the reader do your sorting.

The honest-table checklist

Right instrument - do they need exact values or a lookup? If they need a shape, draw it instead.
Precision - round to what the data support; four decimals on a noisy rate is a lie of false confidence.
Order - sort rows by the quantity of interest; the ranking is information.
Trim - summarize to the comparison (decades, top/bottom); don’t dump raw rows.
Labels - human column names and units, so no cell needs a footnote to be understood.

27.5 Tables and Graphics Are Partners

The two instruments are not rivals; a good report uses both. Show the graph for the pattern - “strikeouts have soared” - and put the table beside it for the reader who needs the exact decade numbers to quote in their own argument. The graph earns the reader’s belief; the table arms them with the specifics. Reach for whichever answers the question in front of you, and often, reach for both.

27.6 Challenge

Do One Yourself

Rebuild the decade table but add a column for how much each stat changed from the previous decade. Which display - this table or the small-multiple graph from the last chapter - makes the change clearer, and for which reader?
Take the full 32-team draft data and design the single table you would actually publish. What do you sort by, how many rows do you keep, and to how many digits do you round? Defend each choice in one sentence.
Find a table in a real paper or report with obvious false precision (six decimals on a correlation, say). Round it honestly and note whether any conclusion actually depended on the digits you deleted.

27.7 Where We Go Next

You can now show data honestly - as a picture and as a table - and you have the full toolkit of this book behind you: describe, measure, model, and display. What remains is the hardest thing to teach and the most valuable to learn: judgment, which tool, when, and why. For that, we stop explaining and start showing, walking through real published analyses with every decision laid bare, beginning with the way The Sports Page thinks. Welcome to the apprenticeship.