Example of a Contingency Table

image of SDA contingency table

This contingency table is the cross tabulation of Party Identification (V007) with the presidential candidate the respondent voted for in the 2004 election (V002). People who did not vote for president in 2004 are excluded from the table.

The table demonstrates proper formatting-the independent variable is at the top and percentages total to 100% within each column. This is called percentaging by the independent variable. So, about 97% of Strong Republicans report voting for George Bush, and about 97% of Strong Democrats report voting for John Kerry.

The colors in the table show the likelihood that the cell entries are either smaller or larger than expected-darker blue represents much smaller than expected and darker red represents much larger than expected.

This table also shows one of the odd characteristics of working with weighted data in SDA/DAS. Look at the 'Weak Democrat' column in the table. It shows that 17 Weak Democrats reported voting for George Bush, 95 Weak Democrats reported voting for John Kerry, and no Weak Democrats reported voting for other candidates. If we add 95 plus 17, we should get 112 but the column total for Weak Democrats is 111. Where did the 112th respondent go? The answer has to do wtih how SDA/DAS treats weighted data-it rounds fractional respondents to the next whole number. Rounding is usually discussed in relation to percentages--if there were three cells in a column and each has 33.3% of the total, the column would still add to 100%--but here we will discuss it in relation to respondents.. Weighting involves multiplying each respondent by a weight factor (typically lower or higher than 1.0) and then adding up the totals. SDA/DAS then rounds the cell entries and reports the whole numbers in each cell. So the 17 Weak Democrats who voted for Bush could be as few as 16.6 respondents or as many as 17.4-both would round to 17. And the 95 Weak Democrats who voted for Kerry could be as low as 94.6 or as high as 95.4. If both cells had been rounded up, the column total would be 16.4 plus 94.6 or 111. Properly rounding the respondents in each cell and properly rounding the column total results in something that looks strange, but it really is not.