I want to calculate the prevalence of an attitude among my respondents. I want to arrive at a number where I can state for instance that 20.56 % of my respondent exhibit the attitude.
I made my respondents rate their level of agreement or disagreement with statements regarding their attitudes on a 6-point Likert scale ranging from (1) strongly disagree, (2) disagree, (3) somewhat disagree, (4) somewhat agree, (5) agree and (6) strongly agree.
What is the best statistical approach to use to calculate the percentage prevalence of the attitude among my respondent?
PS: I have seen that some researchers use descriptive statistics such as numerical summaries like mean, standard deviations and P values to explain their data but that does not give a percentage prevalence value. I will like more explanation about this approach as well.
Hello Eric,
The best answer depends on several elements (how many survey questions pertain to the attitude under investigation; how to define "positive," "neutral," or "negative" attitudes operationally; etc.). Ajit's answer also directs you to consider the quality of the ensemble of items.
At the most basic level, however, here's one approach you could use. Imagine that you have a single item for which you'd like to summarize results. For "positive" cases, you could combine those respondents selecting your options 4, 5, and 6 (somewhat agree, agree, strongly agree). Then divide that sum by the number responding to the question--that is the proportion of those endorsing the statement. Multiply by 100 to represent the value as a per cent.
If there are multiple items that are to be summed, you'll need to define a rule for characterizing the result. Such a rule could be: (a) Those selecting 4, 5, or 6 for X% of the items; (b) A total score of Y or higher; or, (c) if you wanted to define the result normatively for your sample, those respondents in the highest Z percentile values of your sample.
David Morgan has a nice set of thread entries for folks dealing with Likert-type scales, you may find it helpful as well: https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
Good luck with your work!
With ordinal regression you estimate the thresholds between response levels of your scale. These are on log-odds scale, which is easily transformed to proportions: Say, theta_2 is the estimated threshold between 2 and 3 on your Likert scale, then the proportion mu_2 of responses y < 3 is:
P(y
Do use R for statistics, there is a package called 'likert' package, which calculates the percentage for each option and give a very nice graph for it.
See in this link how the plots look
http://rcompanion.org/handbook/E_03.html
I'm not sure what you are trying to do, is it just percentages calculations, or you have an outcome which is ordinal and you want to do a model for prediction of this outcome. If so, ordinal regression is used for such models.
Thank you Dr. Hayawi. I am trying to calculate the prevalence of the attitude among the respondents in terms of percentages. Afterwards I intend to use Chi-Square to test relationships and their significance with demographic characteristics, knowledge and perceptions of the respondents
Eric Offei Affedzie I am interested in doing a similar approach to what you have stated - did you figure it out?
I think Eric Offei Affedzie is asking how to infer the proportion of respondents in the population that exhibit the attitude. Maybe one approach is to assume that the employed Likert score from respondents follows a certain distribution (e.g. normal). So, you could try to estimate the related distribution parameters (e.g the mean and the standard deviation). Once you get that, and having defined the minimun score value (Smin) from which you consider that respondents have the attitude, the proportion of population respondents with the attitude is given by 1-P(x
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India