The answer to this question depends on whether you are planning to use the open recall method or the list-based method.
In the open recall method, the enumerator does not read a list of foods or groups to the respondent. Therefore, the number of rows on the questionnaire will not affect responses. However, it is very important not to “collapse” categories in such a way that it is not possible to capture information on the ten distinct Minimum Dietary Diversity for Women of Reproductive Age (MDD W) food groups.
On the model questionnaire, in addition to rows capturing information on the MDD-W food groups, there are six optional and two required categories. It is okay to change the number of rows by dropping the optional categories (e.g. “Sweets”, “Sugar-sweetened beverages”, “Other fats and oils”), but the questionnaire must include the final two rows: a row for foods/ingredients usually used in very small quantities (“Condiments and seasonings” category) and a row for all “other” foods and beverages. Section 2 of Minimum Dietary Diversity for Women – A guide to measurement provides a detailed description of and rationale for each food group and of the other optional and required categories.
In the list-based method, the enumerator does read a list of example foods, in groups. Responses, and the resulting “count” of food groups, are influenced by the total number of categories and by the choices made in disaggregating categories. In general, the larger the number of questions on a list-based questionnaire, the larger the number of “yes” responses, which in some cases leads to a higher count among the ten MDD-W food groups. If users wish to compare across time or space, it is particularly important that the list-based questionnaires remain the same/have the same number of questions.
For both the open recall and list-based questionnaires, it is allowable to add questions to capture information about specific, targeted food items. For the open recall, this will not bias responses. For the list-based questionnaires, additions should be few and made thoughtfully, to avoid biases in responses and in the constructed indicator.
