I am writing a paper for publication about a clinical support program at an adult mental health residential service. The clinical support program was undertaken to review the clinical placement experiences of Undergraduate student nurses in relation to orientation, induction and clinical support structures, and identify the functions and needs of preceptors.
It depends upon your study design and primary outcome, as such no tailor made statistical method can be employed ignoring above mentioned factors.
as suggested by Mr Saroj that it depends upon your study design, I have attached a graph as model may it help you, which is on degradation on natural resources in a study area. In this study we have asked particular question with defined scale of answers.
@James,
Obviously, you might have developed an instrument with a number of parameters with Likert scale that varies from 1 to 5. The first step in a possible strategy would be to serve the same instrument with reshuffled questions to the subjects and record their response against the previous ones. Using psychometric methods advisably True Score Method calculate weights and apply them to original scores not only covering them into continuous values but also giving weights according to the reliability level of the observers. The True Score Method or any such method may also be applied in comparison to expert ranking or clinical evaluation against nurse students' evaluation.. However, number of parameters to be evaluated quality of health or improvement of patients should not ideally be less than ten.
Now, the weights thus obtained may be used to rank the nurse students for their performance and placement.
Some simple nuts and bolts:
You have ordinal data, which you can't treat with simple t-tests. You want to use non-parametric statistics (Spearman's rho, WIlcoxon/Kruskal-Wallis, Friedman...)
...or, you could treat the Likert scale as a multinomial outcome, and model it accordingly. An alternative would be, perhaps, a generalized logit model (I'm presuming that proportional odds assumptions won't hold, hence the generalized logit approach).
How many Likert scales do you have, and how many levels do each of them have? 1 to 5? 1 to 10? (etc.)
Hi Mary, the student surveys have a 1-5 scale (e.g. Strongly disagree, Disagree, Neutral, Agree, Strongly Agree). Preceptors surveys have Achieved or Not Achieved responses.
Thanks, James. Consider modelling the student surveys as a multinomial, or as an ordinal logistic. The achieved/not achived could be modeled as binary logistic outcomes. Both approaches can include multiple predictors in the model; just be careful to check for collinearity.
Of course, before doing anything else, check the distributions of each of your outcomes, and each of your predictors, to see how the data behave in terms of overall shape, presence of outliers, over- or under-dispersion, and so on.
A method to bring ordinal values to the interval level is to use Thurnstone Transpormation.
One can also simplify and recalculate the results to a binary variable (1= agree (including strongly), 0 = others) and calcuate % agree within the study population. If lots of Likert scales are used, in one survey this can be a practical solution to present the data pragmatically.
That's a good idea, Robert. It could be useful in cases where the scale is 5-point, ranging from strongly disagree to strongly agree. For other scales, though, such as the NRS-1 (a pain scale ranging from 0 to 10, with 0 being no pain and 10 being worst pain imaginable), it can be tricky to decide where the cutpoint should be for deriving a (0, 1) variable.
Your right, I only meant this for 1 to 5 likert scales. I should have mentioned that.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics