Abstract
For test development in the setting of multidimensional item response theory, the exploratory and confirmatory approaches lie on two ends of a continuum in terms of the loading and residual structures. Inspired by the recent development of the Bayesian Lasso (least absolute shrinkage and selection operator), this research proposes a partially confirmatory approach to estimate both structures using Bayesian regression and a covariance Lasso within a unified framework. The Bayesian hierarchical formulation is implemented using Markov chain Monte Carlo estimation, and the shrinkage parameters are estimated simultaneously. The proposed approach with different model variants and constraints was found to be flexible in addressing loading selection and local dependence. Both simulated and real-life data were analyzed to evaluate the performance of the proposed model across different situations.
Similar content being viewed by others
Notes
Personal communication with the corresponding author.
The normative data were not open access due to “publisher sensitivity issues,” but was obtained from Dr. Paul Barrett through personal communication.
References
Bock, R. D., Gibbons, R., & Muraki, E. (1988). Full-information item factor analysis. Applied Psychological Measurement, 12(3), 261–280. https://doi.org/10.1177/014662168801200305.
Box, G. E. P., & Tiao, G. C. (1973). Bayesian inference in statistical analysis. Reading, MA: Addison-Wesley.
Casella, G., & George, E. I. (1992). Explaining the Gibbs sampler. The American Statistician, 46(3), 167–174. https://doi.org/10.1080/00031305.1992.10475878.
Chen, Y., Li, X., Liu, J., & Ying, Z. (2018). Robust measurement via a fused latent and graphical item response theory model. Psychometrika, 83(3), 538–562. https://doi.org/10.1007/s11336-018-9610-4.
Chen, Y., Li, X., & Zhang, S. (2019). Structured latent factor analysis for large-scale data: Identifiability, estimability, and their implications. Journal of the American Statistical Association,. https://doi.org/10.1080/01621459.2019.1635485.
Chib, S., & Greenberg, E. (1995). Understanding the Metropolis–Hastings algorithm. The American Statistician, 49(4), 327–335. https://doi.org/10.1080/00031305.1995.10476177.
Eaton, M. L. (1983). Multivariate statistics: A vector space approach. Beachwood, OH: Institute of Mathematical Statistics.
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Mahwah, NJ: L. Erlbaum Associates.
Epskamp, S., Rhemtulla, M., & Borsboom, D. (2017). Generalized network psychometrics: Combining network and latent variable models. Psychometrika, 82(4), 904–927. https://doi.org/10.1007/s11336-017-9557-x.
Eysenck, S. B., & Barrett, P. (2013). Re-introduction to cross-cultural studies of the EPQ. Personality and Individual Differences, 54(4), 485–489. https://doi.org/10.1016/j.paid.2012.09.022.
Eysenck, S. B., Eysenck, H. J., & Barrett, P. (1985). A revised version of the psychoticism scale. Personality and Individual Differences, 6(1), 21–29. https://doi.org/10.1016/0191-8869(85)90026-1.
Forero, C. G., & Maydeu-Olivares, A. (2009). Estimation of IRT graded response models: Limited versus full information methods. Psychological Methods, 14(3), 275–299. https://doi.org/10.1037/a0015825.
Gelman, A. (1996). Inference and monitoring convergence. In W. R. Gilks, S. Richardson, & D. J. Spiegelharter (Eds.), Markov Chain Monte Carlo in practice (pp. 131–144). London: Chapman & Hall.
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2004). Bayesian data analysis (2nd ed.). London: Chapman & Hall.
Gelman, A., Meng, X.-L., & Stern, H. S. (1996). Posterior predictive assessment of model fitness via realized discrepancies. Statistica Sinica, 6, 733–807.
Geman, S., & Geman, D. (1984). Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 6(6), 721–741. https://doi.org/10.1109/TPAMI.1984.4767596.
Gilks, W. R., Richardson, S., & Spiegelhalter, D. J. (Eds.). (1996). Markov chain Monte Carlo in practice. London: Chapman & Hall.
Gill, J. (2002). Bayesian methods: A social and behavioral sciences approach. Boca Raton, FL: Chapman & Hall/CRC. https://doi.org/10.1201/9781420057478.
Hastings, W. K. (1970). Monte Carlo sampling methods using Markov chains and their application. Biometrika, 57(1), 97–109. https://doi.org/10.1093/biomet/57.1.97.
Jennrich, R. I., & Sampson, P. F. (1966). Rotation for simple loadings. Psychometrika, 31(3), 313–323. https://doi.org/10.1007/BF02289465.
Jöreskog, K. G. (1969). A general approach to confirmatory maximum likelihood factor analysis. Psychometrika, 34(2), 183–202. https://doi.org/10.1007/BF02289343.
Khondker, Z. S., Zhu, H., Chu, H., Lin, W., & Ibrahim, J. G. (2013). The Bayesian covariance lasso. Statistics and Its Interface, 6(2), 243–259. https://doi.org/10.4310/SII.2013.v6.n2.a8.
Lee, S.-Y. (2007). Structural equation modeling: A Bayesian approach. Hoboken, NJ: Wiley. https://doi.org/10.1002/9780470024737.
Liu, X. (2008). Parameter expansion for sampling a correlation matrix: An efficient GPX-RPMH algorithm. Journal of Statistical Computation and Simulation, 78(11), 1065–1076. https://doi.org/10.1080/00949650701519635.
Liu, X., & Daniels, M. J. (2006). A new efficient algorithm for sampling a correlation matrix based on parameter expansion and re-parameterization. Journal of Computational and Graphical Statistics, 15(4), 897–914. https://doi.org/10.1198/106186006X160681.
Lu, Z. H., Chow, S. M., & Loken, E. (2016). Bayesian factor analysis as a variable-selection problem: Alternative priors and consequences. Multivariate Behavioral Research, 51(4), 519–539. https://doi.org/10.1080/00273171.2016.1168279.
Meng, X.-L. (1994). Posterior predictive p-values. Annals of Statistics, 22(3), 1142–1160. https://doi.org/10.1214/aos/1176325622.
Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H., & Teller, E. (1953). Equations of state calculations by fast computing machine. The Journal of Chemical Physics, 21(6), 1087–1092. https://doi.org/10.1063/1.1699114.
Muthén, B., & Asparouhov, T. (2012). Bayesian structural equation modeling: A more flexible representation of substantive theory. Psychological Methods, 17(3), 313–335. https://doi.org/10.1037/a0026802.
Muthén, B. O., & Asparouhov, T. (2002). Latent variable analysis with categorical outcomes: Multi-group and growth modeling in Mplus. Mplus web notes: No. 4. Retrieved January 20, 2020, from http://www.statmodel.com/download/webnotes/CatMGLong.pdf.
Muthén, L. K., & Muthén, B. O. (1998–2015). Mplus user’s guide (7th ed.). Los Angeles, CA: Muthén & Muthén.
Pan, J., Ip, E. H., & Dubé, L. (2017). An alternative to post hoc model modification in confirmatory factor analysis: The Bayesian lasso. Psychological Methods, 22(4), 687–704. https://doi.org/10.1037/met0000112.
Park, T., & Casella, G. (2008). The Bayesian lasso. Journal of the American Statistical Association, 103(482), 681–686. https://doi.org/10.1198/016214508000000337.
Plummer, M., Best, N., Cowles, K., & Vines, K. (2006). CODA: Convergence diagnosis and output analysis for MCMC. R News, 6, 7–11.
R Development Core Team. (2010). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing.
Reckase, M. D. (1985). The difficulty of test items that measure more than one ability. Applied Psychological Measurement, 9(4), 401–412. https://doi.org/10.1177/014662168500900409.
Reckase, M. D. (2009). Multidimensional item response theory. New York, NY: Springer. https://doi.org/10.1007/978-0-387-89976-3.
Sun, J., Chen, Y., Liu, J., Ying, Z., & Xin, T. (2016). Latent variable selection for multidimensional item response theory models via \(L_{\rm 1}\) regularization. Psychometrika, 81(4), 921–939. https://doi.org/10.1007/s11336-016-9529-6.
Tanner, M. A., & Wong, W. H. (1987). The calculation of posterior distributions by data augmentation (with discussion). Journal of the American Statistical Association, 82(398), 528–540. https://doi.org/10.1080/01621459.1987.10478458.
Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B. Methodological, 58(1), 267–288. https://doi.org/10.1111/j.2517-6161.1996.tb02080.x.
Wang, H. (2012). Bayesian graphical lasso models and efficient posterior computation. Bayesian Analysis, 7(4), 867–886. https://doi.org/10.1214/12-BA729.
Wirth, R. J., & Edwards, M. C. (2007). Item factor analysis: Current approaches and future directions. Psychological Methods, 12(1), 58–79. https://doi.org/10.1037/1082-989X.12.1.58.
Yuan, M., & Lin, Y. (2007). Model selection and estimation in the Gaussian graphical model. Biometrika, 94(1), 19–35. https://doi.org/10.1093/biomet/asm018.
Zhang, S., Chen, Y., & Li, X. (2019). mirtjml: Joint maximum likelihood estimation for high-dimensional item factor analysis. Retrieved January 20, 2020, from https://cran.r-project.org/web/packages/mirtjml/index.html.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Derivation of Conditional Distributions
Equation (2) can be rewritten as \(y_{ij} ={{\varvec{\uplambda }}}_{j}^\mathrm{T} {\varvec{\uptheta }}_{i} +\varvec{\upmu }_{j} +\varvec{\upvarepsilon }_{ij} \). After rearranging the loading vector for item j, the trait vector is \({\varvec{\uptheta }}_{i} =\left( {{\begin{array}{*{20}c} {{\varvec{\uptheta }}_{i}^{\prime } } \\ {{\varvec{\uptheta }}_{i}^{\prime \prime } } \\ \end{array} }} \right) \), where the \(K' \times 1\) vector \({\varvec{\uptheta }}_{i}^{\prime } \) and \(K'' \times 1\) vector \({\varvec{\uptheta }}_{i}^{\prime \prime } \) are associated with the unspecified loadings and zero/specified loadings, respectively. Similarly, one can denote the rearranged trait matrix as \(\varvec{\Theta }=\left( {{\begin{array}{*{20}c} \varvec{\Theta }^{\prime } \\ \varvec{\Theta }^{\prime \prime } \\ \end{array} }} \right) =\left( {{\begin{array}{*{20}c} {{\varvec{\uptheta }}_{1}^{\prime } , \ldots {\varvec{\uptheta }}_{i}^{\prime } , \ldots {\varvec{\uptheta }}_{N}^{\prime } } \\ {{\varvec{\uptheta }}_{1}^{\prime \prime } , \ldots {\varvec{\uptheta }}_{i}^{\prime \prime } , \ldots {\varvec{\uptheta }}_{N}^{\prime \prime } } \\ \end{array} }} \right) \). Denote y\(_{i(-j)}\) as the subvector of y\(_{i}\) with the jth element deleted. The conditional distribution for \({\varvec{\uplambda }}_{j}^{\prime } \) can be obtained as:
With \(\psi _{jj}^{*} =\psi _{jj} -{\varvec{\uppsi }}_{j}^\mathrm{T} {\varvec{\Psi }}_{-jj}^{-1} {\varvec{\uppsi }}_{j} \), \(\mathbf{Y }_{j}^{*} =\mathbf{Y }_{j} -{\varvec{\uppsi }}_{j}^\mathrm{T} {\varvec{\Psi }}_{-jj}^{-1} \mathbf{Y }_{-j} \) or \(y_{ij}^{*} =y_{ij} -{\varvec{\uppsi }}_{j}^\mathrm{T} {\varvec{\Psi }}_{-jj}^{-1} \mathbf{y }_{i(-j)} \), and \({\varvec{\upmu }}_{j}^{*} =\mu _{j} -{\varvec{\uppsi }}_{j}^\mathrm{T} {\varvec{\Psi }}_{-jj}^{-1} ({\varvec{\upmu }}_{-j} +{\varvec{\Lambda }}_{-j} {\varvec{\Theta }})\) or \(\mu _{ij}^{*} =\mu _{j} -{\varvec{\uppsi }}_{j}^\mathrm{T} {\varvec{\Psi }}_{-jj}^{-1} ({\varvec{\upmu }}_{-j} +{\varvec{\Lambda }}_{-j} {\varvec{\uptheta }}_{i} )\), it becomes:
With \({\varvec{\uplambda }}_{j}^{\prime } \sim {N}(\mathbf{0 }, \mathbf{D }_{\tau _{j} } )\), one has:
The conditional distribution for \(\tau _{jk} \) can be expressed as:
The conditional distribution for \(\delta _{j}\) can be expressed as:
Similarly, with \({\varvec{\uplambda }}_{j}^{"} \sim {N}(\lambda _{0j} , \mathbf{H }_{0j} )\), the conditional distribution for \({\varvec{\uplambda }}_{j}^{"} \) can be obtained as:
The conditional distributions \(p(\varvec{\uptheta }_{i}|\mathbf{Y} , \varvec{\Lambda }, \varvec{\upmu }, \varvec{\Phi }, \varvec{\Psi })\) and \(\left. p(\varvec{\upmu } \right| \mathbf{Z }, {\varvec{\Lambda }}, {\varvec{\Phi }}, {\varvec{\Psi }})\) are similar to those presented in Lee (2007, pp. 146–147), as:
where \(\mathbf{V }=\sum \nolimits _{i=1}^N {(\mathbf{y }_{i} -{\varvec{\Lambda } \varvec{\uptheta }}_{i} )} \).
For PCIRM-LI, \(\varvec{\Psi } =\text {diag}(\psi _{jj})\) is modeled as a diagonal matrix. The conditional distribution for \(\psi _{jj}\) can be expressed as: \(\left. {p(\psi _{jj}^{-1} } \right| \mathbf{Z }, {\varvec{\Lambda }}, {\varvec{\upmu }}, {\varvec{\Phi }})\propto p(\mathbf{y }_{j} \vert \mathbf{y }_{-j} , {\varvec{\Lambda }}, {\varvec{\Theta }}, {\varvec{\upmu }}, {\varvec{\Phi }})p(\psi _{jj}^{-1} )\). It can be further simplified as:
Block Gibbs Sampler to Draw \(\Sigma \) and \(\Psi \) with Local Dependence
The conditional distribution \(\left. p({\varvec{\Sigma }} \right| \mathbf{Y }, {\varvec{\upmu }}, {\varvec{\Theta }}, {\varvec{\Lambda }}, {\varvec{\Phi }}, {\varvec{\uptau }}_{s} , \delta _{s} )\) can be decomposed as follows:
where \({\varvec{\uptau }}_{s} =(\tau _{ij} )_{i<j} \) is the vector of the latent scale parameters, and
For \(j = 1, \ldots , J\) and without loss of generality, one can partition and rearrange the columns of \(\varvec{\Sigma }\) and S as follows:
where \(\sigma _{jj}\) is the jth diagonal element of \(\varvec{\Sigma }\), \(\varvec{\sigma }_{j}=\left( \sigma _{j1}, {\ldots , }\sigma _{j,j-1}, \sigma _{j,j+1}, \ldots , \sigma _{jJ} \right) ^{\mathrm {T}}\)is the vector of all off-diagonal elements of the jth column, and \(\varvec{\Sigma }_{-jj}\) is the (\(J \quad -\) 1) \(\times \) (\(J \quad -\) 1) matrix resulting from deleting the jth row and jth column of \(\varvec{\Sigma }\). Similar, \(s_{-jj}\) is the jth diagonal element of S, s\(_{j}\) is the vector of all-diagonal elements of the jth column of \(\mathbf{S }\), and \(\mathbf{S} _{-jj}\) is the matrix with thejth row and jth column of S deleted. Then we have:
where \(\mathbf{M} _{\tau }\) is the diagonal matrix with diagonal elements \(\tau _{j1}, \ldots , \tau _{j,j-1}, \tau _{j,j+1}, \ldots , \tau _{jJ}\).
Let \(\varvec{\beta }=\varvec{\upsigma }_{j}\) and \(\varvec{\gamma }=\sigma _{jj}-\varvec{\upsigma }_{j}^{\mathrm {T}}\varvec{\Sigma }_{-jj}^{-1}\varvec{\upsigma }_{j}\). It can be shown that:
After drawing from the above conditional distributions, we can obtain \(\varvec{\sigma }_{j}=\varvec{\beta }\), \(\varvec{\sigma }_{j}^{\mathrm {T}}=\varvec{\beta }^{\mathrm {T}}\) and \(\sigma _{jj}=\varvec{\gamma }+\varvec{\sigma }_{j}^{\mathrm {T}}\varvec{\Sigma }_{-jj}^{-1}\varvec{\sigma }_{j}\), then the jth column and row of \(\varvec{\Sigma } \) can be updated one at a time. At the end, \({\varvec{\Psi }}={\varvec{\Sigma }}^{-1}\)is computed.
The conditional distribution for \({\varvec{\uptau }}_{s} =(\tau _{ij} )_{i<j} \) can be expressed as:
It can be shown that for \(i < j\),
The conditional distribution for \(\delta _{s}\) can be expressed as:
Details of Nonzero Loading Estimates
The Revised Eysenck Personality Questionnaire Short Scale
No. | Content | T |
---|---|---|
1 | Would you take drugs which may have strange or dangerous effects? | P |
2 | Do you prefer to go your own way rather than act by the rules? | P |
3 | Do you think marriage is old fashioned and should be done away with? | P |
4 | Do you think people spend too much time safeguarding their future with savings and insurance? | P |
5 | Would you like other people to be afraid of you? | P |
6(R) | Do you take much notice of what people think? | P |
7(R) | Would being in debt worry you? | P |
8(R) | Do good manners and cleanliness matter much to you? | P |
9(R) | Do you enjoy co-operating with others? | P |
10(R) | Does it worry you if you know there are mistakes in your work? | P |
11(R) | Do you try not to be rude to people? | P |
12(R) | Is it better to follow society’s rules than go your own way? | P |
13 | Are you a talkative person? | E |
14 | Are you rather lively? | E |
15 | Can you usually let yourself go and enjoy yourself at a lively party? | E |
16 | Do you enjoy meeting new people? | E |
17 | Do you usually take the initiative in making new friends? | E |
18 | Can you easily get some life into a rather dull party? | E |
19 | Do you like mixing with people? | E |
20 | Can you get a party going? | E |
21 | Do you like plenty of bustle and excitement around you? | E |
22 | Do other people think of you as being very lively? | E |
23(R) | Do you tend to keep in the background on social occasions? | E |
24(R) | Are you mostly quiet when you are with other people? | E |
25 | Does your mood often go up and down? | N |
26 | Do you ever feel ‘just miserable’ for no reason? | N |
27 | Are you an irritable person? | N |
28 | Are your feelings easily hurt? | N |
29 | Do you often feel ‘fed-up’? | N |
30 | Are you often troubled about feelings of guilt? | N |
31 | Would you call yourself a nervous person? | N |
32 | Are you a worrier? | N |
33 | Would you call yourself tense or ‘highly-strung’? | N |
34 | Do you worry too long after an embarrassing experience? | N |
35 | Do you suffer from ‘nerves’? | N |
36 | Do you often feel lonely? | N |
Note. Item responses were dichotomous (i.e., Yes/No); T \(=\) targeted trait; P \(=\) psychoticism; E \(=\) extraversion; N \(=\) neuroticism; negatively worded items were marked by “R.”
Rights and permissions
About this article
Cite this article
Chen, J. A Partially Confirmatory Approach to the Multidimensional Item Response Theory with the Bayesian Lasso. Psychometrika 85, 738–774 (2020). https://doi.org/10.1007/s11336-020-09724-3
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11336-020-09724-3