I have used tool in the Likert Scale. How can I do scale sensitivity analysis?
Hello Rajan,
If, by sensitivity, you mean trying to find the score (on some scale) which maximizes the ratio of True positives / (True positives + False negatives) with respect to some dichotomous target variable (e.g., survive/did not survive), then there are several paths you could take.
1. First, generate a composite score for the set of related/relevant responses from the scale. Rensis Likert would use a summated score for this purpose. More modern methods, such as: (a) ordinal factor analysis; (b) ordinary factor analysis using polychoric correlations as the basis; or (c) item response theory models for polytomous variables are candidates as well.
Then, perform an AUC/ROC analysis. Check the actual sensitivity for the chosen threshold, ideally with additional data.
2. Use the related/relevant responses from the scale as the independent variables and the dichotomous target value as the dependent variable, and run a logistic regression, treating the IVs as ordinal.
Good luck with your work.
I don't think there really is such a thing. But if you analyse using a Chi square test, it's kind of like a sensitive analysis because it uses expected values :)
Bueno, como tienes 5 opciones: 1 (muy malo); 2(malo); (regular); 4(bueno); y, 5 (muy bueno). Entonces primero aplicas tu encuesta con las 5 categorías; después descartas la controversial categoría 3 (reguilar, o, más o menos). Y comparas resultados con F de Fisher (data cuantitativa) o con chi2 (data cualitativa) y así mediante el con traste de hipótesis puedes medir el grado de sensibilidad con; o, sin, LA CATEGORÍA 3.
También puedes hacer, similarmente, el mismo proceso, pero ahora descartando la(s) categorías 1(muy malo) y 2(malo) en el caso de instituciones empresariales o no empresariales, que se consideran "eficientes" en el mercado. Así, qué ocurre que la empresa eficiente te "exige" que trabajes sin la 1 y la 2; pero tú, para salir de dudas, aplicas todas las categoría, luego descartas la 1; luego la 2, luego la 3. Entonces haces un excelente trabajo de sensibilidad y obtienes resultados integrales que demuestran que la empresa no es tan eficiente como apriori afirman sus ejecutivos y propietarios; obviamente, tu calidad de investigador se revaloriza. Atentamente, [email protected]
Juan Pastor Meza Chavera is it possible to write this response in English?
I just translated Juan Pastor Meza Chavera reply from "Spanish" to "English"
Well, since you have 5 options: 1 (very bad); 2(bad); (regular); 4(good); and, 5 (very good). So first, you apply your survey with the five categories; then, you discard the controversial category 3 (regular, or, more or less). And you compare results with Fisher's F (quantitative data) or chi2 (qualitative data); thus, by contrasting hypotheses, you can measure the degree of sensitivity with or without CATEGORY 3.
You can also do, similarly, the same process, but now discarding the category(ies) 1(very bad) and 2(bad) in the case of business or non-business institutions, which are considered "efficient" in the market. So, what happens is that the efficient company "demands" you to work without 1 and 2; but you, to get rid of doubts, apply all the categories, then discard 1; then 2, then 3. So you do an excellent job of sensitivity, and you get comprehensive results that show that the company is not as efficient as its executives and owners a priori claim; obviously, your quality as an investigator is revalued.
Gracias Bijay, Sólo hablo español. Sorry. Pero así nos apoyamos. Saludos.
- Badges
- Science topic