Hello all,
Thanks a lot for taking out time to read and respond to the question.
So, I have designed a behavioral usage survey with a question that captures technology use (independant variable). In the pilot phase, likert scale options had the following values:
Never, A few times a year, Once a month, A few times a month, Once a week, A few times a week, Every day
However, to capture more specific responses, I converted the likert scale to the following.
- » Never used
- » Less than 1 hour
- » 1-2 hours
- » 3-5 hours
- » 6-9 hours
- » 10-19 hours
- » 20-29 hours
- » 30-39 hours
- » 40 hours or more
The mistake I made was that I had also translated the survey (eventually decided to go with only English survey), but I forgot to turn it off initially.
The bad news is that I have significant answers (45%) that in the old scale (i.e. never, a few times a year) and the rest in the new scale.
Good news is that it is an issue only with this independent variable as this is the only question I changed after pilot study. SO, that rest of the translated questions map accurately to the old survey.
My question : Is there anyways I can save the 45% of the data for this particular independent variable question? If so, what is the best method? And how can I talk about this in the paper?
You could convert your Likert-scored wording into estimates of hours per week.
But first, you should test whether there are significant differences between the two types of respondents. You can assess this via t-Tests on relevant variables, where you use the type of response as your two-category variable. If the two categories of respondents are equivalent, then you can go on to compare the correlations you get with the two versions of your variable.
Hi David L Morgan Thanks a lot for the answer. Do I still need to do the test if the respondents who answered the translated version and the English version belong to the same population of teachers?
Because, ideally what I want to do is what you suggested convert my likert wordings into hours per week so that I can use it for regression tests with dependent variables (without having to throw out out those respondents who asked translated version of the survey)
I think you do need to determine if the respondents come from the same population, in the sense that they do not differ on key variables (e.g., gender, number of years of experience, etc.)
The reason is that how your two measures relate to other variables might be tied to who responded to each, rather than how each measured your key concept.
I am doubtful there is a defensible way to map the categories of times per year onto the categories of hours per week (or vice versa). The two versions of the question call for radically different cognitive processing by the respondent, and therefore are essentially measuring two different things.
How do the distributions compare in your data? I'm guessing that times per year is top-heavy, with most responses being in the weekly to daily categories, whereas the hours per week could be more bell-shaped, with only a few in the very highest categories of that version. If so, then it wouldn't make sense to map high categories on one onto the high categories of the other, middle to middle, and low to low. Maybe you could say "never" means the same on both -- provided that there is roughly the same % saying never on both versions of the question. If not, then even that partial mapping would be highly suspect.
You might do an after-the-fact "pilot" study, in which some new respondents are asked to complete a survey that includes both versions of the question (separated by some other questions). Then you would at least have some empirical basis for deciding which categories of one version map most consistently onto which categories of the other version. Or the results might convince you that they do not consistently map in any particular way, in which case you might have to drop the 45% of your final sample that got the wrong version.
Or, perhaps you will have to settle for dichotomizing both versions of the question, using only 2 categories: "Very Little" (maybe never through a few times a month and never through 1-2 tours per week) versus "Considerably" (all other categories of response). Comparing the distributions of the two versions of the question in your data could give you some guidance on where to put the cut-points. You would want the cumulative % of all categories included under "Very Little" to be approximately the same in the two versions of the question.
thanks a lot David L Morgan and Burke D. Grandjean for your answers! Really appreciate your help.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey