Loevinger's H coefficients and non parametric Item Responses Models
loevh varlist [, pairwise pair ppp pmm noadjust generror(newvar) replace graph monotonicity(string) nipmatrix(string)]
varlist is a list of two or more existing dichotomous ou polytomous variables.
Description
loevh allows verifying the fit of data to the Monotonely Homogeneous Mokken Model or to the Doubly Monotone Mokken Model. It computes the Loevinger H scalability coefficients, and several indexes in the field of the Non parametric Item Response Theory.
Options
pairwise. By default, all the individuals with one or more missing values are omitted. pairwise allows to use the complete information by pair of items.
pair displays statistics and the value of the Loevinger H coefficient for each pair of items.
ppp displays the P++ matrix.
pmm displays the P-- matrix.
noadjust approximates the tests statistics like the MSP software (Molenaar et al. (2000)).
generror creates a new variable containing the number of Guttman errors produced by each individual.
replace allows replacing the variable defined by the generror option.
graph displays graphs (only with the ppp, pmm and generror options).
monotonicity displays indexes in order to check the monotonicity of the data (Monotone Homogeneity Mokken Model). This option produces outputs similar to the MSP software. The string contains several suboptions: minvi, minsize, siglevel and details. If you want use all the default values, type *. minvi defines the minimal size of a violation of monotonicity (0.03 by default) minsize defines the minimum size of groups of patients to check the monotonicity (by default, the number of individuals divided by 10 with more than 500 individuals, the same number divided by 5 with more than 250 individuals, and the same number divided by 3 for a smaller number, with a minimum of 50) siglevel defines the significance level for the tests (0.05 by default) details displays more details with polytomous items
nipmatrix display indexes in order to check the non-intersection (Doubly Monotone Mokken Model). This option produces outputs similar to the MSP software. The string contains several suboptions: minvi and siglevel. If you want use all the default values, type *. minvi defines the minimal size of a violation of non-intersection (0.03 by default) siglevel defines the significance level for the tests (0.05 by default)
Remarks
For detailed informations on the Loevinger's H coefficients, see Loevinger (1948) or Hemker and al. (1995). For details about the analysis of non parametric Mokken models, see for example the MSP 5.0 manual.
Example
. loevh itemA1-itemA7
. loevh itemA*, pair monotonicity(*) ppp pmm nipmatrix(minvi(0.05) siglevel(0.01))
. loevh item*, pairwise generror(error) graph
Results
The Loevinger's H coefficients between all the pairs of items, for each item with respect of all the others items and for the set of items are respectively saved in the matrices r(loevHjk), r(loevHj) and in the scalar r(loevh).
The empirical Guttman errors between all the pairs of items, associated to each item and relied to the scale are respectively saved in the matrices r(eGuttjk), r(eGuttj) and in the scalar e(Gutt).
The theorical Guttman errors between all the pairs of items, associated to each item and relied to the scale are respectively saved in the matrices r(eGuttjk0), r(eGuttj0) and in the scalar e(Gutt0).
The values of the Z statistics and the corresponding p-values associated to the Loevinger H coefficients are respectively saved in the matrices r(zHjk), r(pvalHjk), r(zHj), r(pvalHj) and in the scalars e(zH) and r(pvalH).
The P++ and P-- matrices are saved in r(ppp) and r(pmm).
The used number of individuals per items pair is saved in r(Obs).
References
Hemker B. T., Sijtsma K. and Molenaar I. W., Selection of unidimensional scales from a multidimensional item bank in the polytomous Mokken IRT model, Applied Psychological Measurement, vol.19(4), 1995, pp. 337-352.
Loevinger J., The technique of homogeneous tests compared with some aspects of "scale analysis" and factor analysis. Psychological bulletin, vol. 45, 1948, pp. 507-530.
Molenaar I. W., Sijtsma K. and Boer P. MSP5 for Windows - User's Manual, 2000, 105 pages.
Author
Jean-Benoit Hardouin, PhD, assistant professor EA 4275 "Team of Biostatistics, Clinical Research and Subjective Measures in Health Sciences" University of Nantes - Faculty of Pharmaceutical Sciences 1, rue Gaston Veil - BP 53508 44035 Nantes Cedex 1 - FRANCE Email: jean-benoit.hardouin@univ-nantes.fr Websites AnaQol and FreeIRT
Also see
Online: help for traces, msp, gengroup, mokken if installed.