I made an error in my qualtrics survey, and used a 5-point Likert scale, when the measure was supposed to use a 4-point scale. I have collected a lot of my data for my dissertation, and am now very nervous. Do I have to go back and collect all of my data again?
-
1$\begingroup$ Welcome to CogSci. Why do you think that matters? What analysis are you going to use? In short, please add backgrounds as the question is unclear. $\endgroup$– AliceD ♦Sep 18, 2016 at 12:40
1 Answer
In general, from my extensive experience, using a four or a five point response scale is not going to change much the psychometric properties of a typical psychological self-report scale (e.g., reliability and factor loadings). I also imagine that if you were to measure a multi-item scale with a four point version and a five point version that the correlation would be very close to the test-retest correlation you get with the five point scale.
More extreme changes (e.g., using a 2-point instead of a 5-point) or radically altering the response options would eventually start to create issues, but in the 4 to 7 point range, I think the effects would be subtle, especially if you are dealing with a multi-item scale.
The main issue you will have is that you wont be able to compare your descriptive statistics (e.g., means and standard deviations) with other authors who have used the standard 5-point scale. Whether this is an issue, depends on the aims of your analyses. If you are focused on relationships between variables (e.g., correlations, regressions, group differences, SEM, and so on), then whether you use a 4 or a 5-point scale is not going to make much difference. If you want to compare the means of your sample to the means of previous samples, or if there are particular scoring rules that involve the original metrics (e.g., some tests have cut-off scores used for classification), then you have more of a problem.
There are ways to approximate a rescaling. I.e., estimating what the means and SDs would have been had you used the 5-point instead of the 4-point scale.
rescaled_old_score = (old_score - OLD_MIN) / (OLD_MAX - OLD_MIN)
new_score = rescaled_old_score * (NEW_MAX - NEW_MIN) + NEW_MIN
Here's what the function looks like in R:
rescale_item <- function(old_score, old_min = 1, old_max = 4, new_min = 1, new_max = 5) {
rescaled_old_score <- (old_score - old_min) / (old_max - old_min)
rescaled_old_score * (new_max - new_min) + new_min
}
template <- data.frame(original = 1:4)
template$new <- round(rescale_item(template$original, 1, 4, 1, 5), 2)
template
which leads to the following rescaling:
original new
1 1 1.00
2 2 2.33
3 3 3.67
4 4 5.00
Of course, this is only an approximation. Better estimates of the mapping values could be obtained by administering the tests using both scales to the same sample.
-
$\begingroup$ Jeromy, THANK YOU. This has been an unbelievable help. I hugely appreciate it!! $\endgroup$– JHKSep 20, 2016 at 10:07