Tables are effective tools for telling a compelling story if they are made simple. But the software used to create tables adds unnecessary and distracting packaging - the lines and labels used to interpret the data - and analysts do not always organize tables in a way that makes intuitive sense.
Assume that you are trying to show that the way beer is displayed facilitates beer thefts from stores. Some stores display the beer near the front entrance and some at the rear. You are trying to show that rear display stores have fewer incidents of beer thefts than those where the display is in the front. Table 1 gets in the way of this message. The data are poorly organized and the packaging is distracting.
Table 2 properly organizes the data. The percentages are made central to the story. Because the raw numbers do not tell the main story, but may be useful to a reader who wants to look more closely, they are made subservient by enclosing them in brackets. Finally, instead of row percentages (as in the first table), Table 2 uses column percentages.
Whenever we examine a relationship in which something may be causing something else, it is best to put the cause in the columns and use column percentages. Then, make comparisons across the rows. Here, we see immediately that 29 percent of the front-display stores had no thefts compared to almost 83 percent of the rear-display stores. At the opposite extreme, almost 46 percent of the front-display stores had three or more thefts, but none of the rear-display stores did.
Table 2 has less packaging. The bold borders have been removed and replaced by thin lines. Inside, the only remaining line separates the title from the content. Instead of lines, space is used to guide the reader's eye across rows and down columns. By informing the reader in the title that the important numbers are percentages (and the raw numbers are in brackets), there was no need to include a percent symbol in each cell. Only the column total remains. This tells the reader that the important sum is vertical. Finally, all percentages are rounded to one decimal place, thus allowing the column figures to line up, making interpretation easier. With all of these changes, most of the content of the table is data rather than packaging.
A problem often has multiple causes. Though tables can be constructed to show large numbers of causes, a single table communicates poorly when you examine more than two causes. The basic principles of table construction remain the same:
- All the causes go in the same direction (usually columns).
- Summation goes in the direction of the cause (down columns).
- Comparison of causes goes in the opposite direction (across rows, if causes are in columns).
Table 3 is called a three-dimensional table because three things are examined (the earlier tables were two dimensional). Table 3 answers the question: is the relationship between display location and theft different for two different store chains (Drink-Lots and Tippers). The answer is that it is not. There is the same basic pattern for both chains that we saw in Table 2. In both cases, we sum the column and compare front-display stores to rear-display stores. This implies that any store that displays beer at the rear will experience less theft, regardless of which chain it belongs to.
In effect, Table 3 holds constant the type of store. Other factors can be held constant if we think they are important. For example, we could group stores by size (small, medium and large), and separately analyze the relationship between display location and theft for each size category. This would require three panels, but otherwise the same principles apply.
Take note of several other features of Table 3:
- If you add the raw numbers (in brackets) in the Drink-Lots cells to their corresponding cells under Tippers, you get the raw numbers in Table 2. In other words, Table 2 is a summary of Table 3. But you cannot derive Table 3 from Table 2.
- Because Table 3 contains two possible causes of the problem, we have added a vertical line to draw attention to the two types of stores.
- The row labels apply to both store types, so there was no need to duplicate them.
- Because of rounding in the percentages, they sometimes add to over 100. In some instances these sums can be just under 100, usually 99.9. Such small deviations are seldom of much concern.
If you routinely produce the same tables for the same decision-makers, show them several different table formats with the same data. Determine which format helps them the most, and then use this standard format.
Table 1: Location and Beer Theft (June)
|Location of display|
|Number of theft reports||Front||Rear||Total|
|0||7 (17.5%)||33 (82.5%)||40|
|1-2||6 (46.15%)||7 (53.85%)||13|
|3 or more||11 (100%)||0 (0%)||11|
|Total||24 (37.5%)||40 (62.5%)||64|
Table 2: Percent of Stores with Reported Beer Thefts (Numbers in Brackets)
|Location of display|
|Thefts in June||Front||Rear|
|0||29.2 (7)||82.5 (33)|
|1-2||25.0 (6)||29.2 (7)|
|3 or more||45.8 (11)||0.0 (0)|
|Total||100 (24)||100.1 (40)|
Table 3: Percent of Stores with Reported Beer Thefts by Retail Chain (Numbers in Brackets)
|Drink-Lots Stores||Tippers Stores|
|Thefts in June||Front Display||Rear Display||Front Display||Rear Display|
|3 or more||46.2||(6)||0.0||(0)||45.5||(5)||0.0||(0)|