Newsletter 12 / 2013
|Upcoming Events 2014||New Neusrel Software-Features||New Customers about Newsrel|
New Neusrel Features
- Latent Factor Exploration for causal models:
Do you know this? You have surveyed many items and attributes and you have an idea how they build up to form latent variables. However your model shows bad fit measures and doubts are growing about your measurement model. Then you try exploratory principal component analysis but the results are not satisfying or they do not result in latents that produce structural model with higher explanation power.In this situation you wish to have a method that explores latent factors out of existing attributes with regard to maximize explanatory power of your structural model. And this method is now available within the NEUSREL platform. It employs Universal Multi-Target Regression to find components that maximizes explanatory power. In step two components are converted into a measure similar to factor loadings and will then be rotated. The users part is then to interpret results with domain knowledge and to finally define new factors. Practical applications show that the new method results in factors with not only higher explanatory but also higher face validity.
- Dealing with Omitted Variables – The new Two-Stage-Least-Squares feature:
One of the main limits for the validity of causal models are the absence of important variables that have influence on several other variables in the model. In this case the model suggests that a variable in the model has an impact which in truth steams from the omitted variable. This is why we strive for holistic model.
The 2SLS approach is designed to eliminate the variance of the omitted variable out of the model variables. It does this by the use of variables that fulfil two properties: First they are predictive to the included variables. Second, it can be assumed that they itself are not dependent from the omitted variable. This is true for natural facts such as many demographic properties. NEUSREL now allows correcting your data with a nonlinear 2SLS approach.