Check all that apply: multiple responses

In brief

In a question like “Which media do you use? Check all that apply”, each option sits in its own column (checked = 1, otherwise 0). Under “Describe columns” you combine these columns into a set; Statable suggests columns such as media_1 to media_6 by itself. The set then appears in the question “How is something distributed?” (frequencies) and in the question about groups (“Does something occur at different rates in different groups?”, as “What is counted?”).

Reading the result

• For each option: the number of people who named it, the percent of cases (share of people) and the percent of responses (share of all checks).
• Percentages of cases add up to more than 100% because each person can name several options. Percent of responses add up to 100%.
• The bar chart shows percent of cases, sorted by frequency; the order of the columns can be chosen.
• With a group column the crosstab shows column percentages: what share of the cases in each group named the option.

Requirements and limits

• Valid cases are those with at least one check (as in SPSS). People who checked nothing are missing from the base. “Include cases without a check” includes everyone with an answer to at least one option - useful when “none of these” is a real answer.
• The checks of one person are not independent. Statable therefore does not calculate a chi-square test for multiple responses; differences between groups are described in percentage points.
• With fewer than 30 valid cases the percentages are uncertain; Statable shows a note.
• Not offered: sets made of columns with codes (“mention 1, 2, 3”) and the set as a group column.

Media used: television n = 24 (77.4% of cases); radio n = 17 (54.8% of cases); N = 31 cases, 72 responses

For specialists

Own code, because PSPP 2.1.3 reads MRSETS but does not know MULT RESPONSE and CTABLES does not support multiple response sets. Calculated as SPSS MULT RESPONSE for a multiple dichotomy (counted value, e.g. 1): a case is valid if at least one column has the counted value; percent of cases = count / valid cases, percent of responses = count / all responses. Checked against a calculation in R (colSums). A test of independence is left out: multiple responses need methods for marginal independence (Agresti and Liu 1999; Bilder and Loughin 2004, bootstrap or Rao-Scott), which Statable does not offer.

Typical examples

• "Which media do you use?" (several can be checked): percent of cases per medium.
• Media used by gender: column percentages per group.