lavaan.survey
An R package allowing for complex sampling analysis of SEM in R using the lavaan and survey packages. Uses the Satorra-Bentler approach to multivariate complex sample analysis described in Muthén & Satorra (1995, Sociological methodology).
Thanks to the features written by Thomas Lumley in the survey package, lavaan.survey has features novel to SEM. For example, it allows for a finite population correction, and you can do GREG estimation of SEM models.
- The Journal of Statistical Software paper (2014) contains all examples and technical background.
- Download examples and data from the paper
SQP 2.0
- Won the 2014 Warren J. Mitofsky Innovators Award from the American Association of Public Opinion Research (AAPOR)
- Described in the Wiley book “Design, Evaluation, and Analysis of Questionnaires for Survey Research, Second Edition” (Saris & Gallhofer 2014)
Registration and use of the program is open to all and free at http://sqp.upf.edu/.
Jrule for Mplus
A standalone Windows program that reads in Mplus output with modification index and expected parameter change information and provides a user interface to taking into account the power of the score test as described by Saris, Satorra & Van der Veld (2009, Structural Equation Modeling).
“Jrule for AMOS”
(To use this Excel sheet you will need to install the “Real Statistics” Excel add-on.)
SQP 1.0
The old Windows version of SQP, from 2005. SQP is a program for the prediction of reliability and method effect of a survey question. The prediction is based on characteristics of the question, such as its linguistic complexity, the number of scale points, presence/absence of a “don’t know” option, etc.
Available for free at http://prod.sqp.nl/media/sqp-1.0/.
Differences with SQP 2.0:
- The old program is based on 87 MTMM experiments from 1991–2002, while the new program is based on those results plus the results from the Eureopean Social Survey (ESS) experiments in the 2002–2006 rounds, allowing more countries, languages, and characteristics;
- The old program can in theory give predictions above 1 or below 0 for reliability/method effect coefficients. Though this does not happen often in practice. The new program’s predictions always stay in the allowed range;
- The new program takes into account some interaction effects of the characteristics;
- The new program gives prediction intervals on the quality predictions
- The new program uses more characteristics to predict the quality;
- The new program has lots more features, among them automatic coding of the number of syllables and (for some languages) nouns;
- The old program is a Win32 exe, the new is a web app.
Here is a blurry video of me demonstrating the program: