A scale such as 1-Frequently, 2-Occasionally, 3-Rarely, 4-Never is useful but responses can't be quantified, i.e. define frequently.
Consider requesting a number (open ended) rather than Likert response. If Likert is needed, why not something like "0 = never, 1 = once a week, 2 = twice a week,...7 = seven or more times a week"? Of course, depends on what you are measuring. You can categorize ranges of values as well, "0 = never, 1 = 1 to 4 times a week, ...." But then, I would just have asked for the number (you can always collapse them into categories afterwards).
Also, consider specifying a time frame within the question itself, (e.g., "In the past week, how often....?"). Depends on what your measuring; a low frequency behavior may have a longer time frame (e.g., houses bought in past year); a high frequency behavior a shorter time frame (e.g., cigarettes smoked yesterday). Consider keeping the time frame as short as needed to avoid recall bias.
Best,
Jason
Consider requesting a number (open ended) rather than Likert response. If Likert is needed, why not something like "0 = never, 1 = once a week, 2 = twice a week,...7 = seven or more times a week"? Of course, depends on what you are measuring. You can categorize ranges of values as well, "0 = never, 1 = 1 to 4 times a week, ...." But then, I would just have asked for the number (you can always collapse them into categories afterwards).
Also, consider specifying a time frame within the question itself, (e.g., "In the past week, how often....?"). Depends on what your measuring; a low frequency behavior may have a longer time frame (e.g., houses bought in past year); a high frequency behavior a shorter time frame (e.g., cigarettes smoked yesterday). Consider keeping the time frame as short as needed to avoid recall bias.
Best,
Jason
Avoid response models with central values (odd categories of responses like 1-5, 1-7 and so on). They create a central tendency sub-sample that might generate a huge noise on your data. Use instead an even number of categories. About quantifiable response anchors, I'm not sure what do you mean. If you mean by raw time basis information, people usually won't be able to give you a good answer. Best bet is on approximations. I study about media usage and that type of thing usually backfires. You might also think of using magnitude estimations, which can use anchors. There's a cheap book about that on amazon from sage green book series.
Still, I might not be understanding what you really want...
This strikes me as just the same as any other Likert-scored data, so if you have multiple questions about the same theoretical construct, you can combine them to create a scale.
Since there have been so many questions here about Likert scales, I have complied a set of resources on this topic:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
I second Jason Hurwitz's response, especially the "it depends" part. It is a very good idea to ask people to answer questions about frequency and also very good to set a time frame. But what response options to offer and what time frame will depend on the behavior (service use) you are trying to measure. You might make different choices to get the discrimination/precision of the options at the point in the frequency count that provides the most useful info. For example, if you are trying to measure use of a service that most people use one, but never return to, you want response options for 0, 1, and 2 to be distinctive options. If you are trying to measure something like video games that some people use a little (1-3 times per week), some use a moderate amount (6 to 10 times per week), and some use a lot (20 to 40 times per week), you may want to just ask "how many times?" and get a frequency report or structure the response options to allow you to categorize people in a meaningful way.
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design