### Interpretation of Parameter Estimates for Fully connected Linear Regression Models

#### Abstract

This article is devoted to the study of interpretation questions of parameter estimates for fully connected linear regression models. In such models, all observed variables contain errors, and true variables are interconnected by linear functional dependencies. A special case of fully connected regression is the well-studied Deming regression. Previously, a weighted total least squares method was used to estimate fully connected regressions. In this article, it is established that the estimates of fully connected linear regression obtained by this method coincide with the estimates of the maximum likelihood method. It was found that it is impossible to interpret fully connected regressions by analogy with multiple regressions, since the former are construct on the assumption that all variables are strongly correlated with each other. A theorem is proved according to which a simultaneous increase in the values of the observed variables in the estimated model of a fully connected linear regression by certain values leads to an increase in the estimates of the true values of the variables by the same values. Using this fact, any model of fully connected linear regression can be interpreted, which is demonstrated by the example of modeling such macroeconomic indicators of the Irkutsk region as the turnover of wholesale and retail trade, as well as agricultural products.

#### Full Text:

PDF (Russian)#### References

Carrasco J.M., Figueroa-Zuñiga J.I., Leiva V., Riquelme M., Aykroyd R.G. An errors-in-variables model based on the Birnbaum–Saunders distribution and its diagnostics with an application to earthquake data // Stochastic Environmental Research and Risk Assessment. 2020. Vol. 34. No. 2. P. 369-380.

Nghiem L.H., Byrd M.C., Potgieter C.J. Estimation in linear errors-in-variables models with unknown error distribution // Biometrika. 2020. Vol. 107. No. 4. P. 841-856.

Han J., Zhang S., Li Y., Zhang X. A general partial errors-in-variables model and a corresponding weighted total least-squares algorithm // Survey Review. 2020. Vol.52. No. 371. P. 126-133.

Bazilevskiy M.P. Metody postroeniya regressionnykh modeley s oshibkami vo vsekh peremennykh. Irkutsk : IrGUPS, 2019. 208 p.

Bazilevskiy M.P. Mnogofaktornye modeli polnosvyaznoy lineynoy regressii bez ogranicheniy na sootnosheniya dispersiy oshibok peremennykh // Informatika i ee primeneniya. 2020. Vol. 14. No. 2. P. 92-97.

Ciccione L., Dehaene S. Can humans perform mental regression on a graph? Accuracy and bias in the perception of scatterplots // Cognitive Psychology. 2021. Vol. 128. P. 101406.

Vicente F.B., Lin D.C., Haymond S. Automation of chromatographic peak review and order to result data transfer in a clinical mass spectrometry laboratory // Clinica Chimica Acta. 2019. Vol. 498. P. 84-89.

Karvonen T., Oates C.J. Maximum likelihood estimation in Gaussian process regression is ill-posed // Journal of Machine Learning Research. 2023. Vol. 24. No. 120. P. 1-47.

Correia S., Guimarães P., Zylkin T. Verifying the existence of maximum likelihood estimates for generalized linear models // arXiv preprint arXiv:1903.01633. 2019.

Bazilevskiy M.P. Issledovanie povedeniya otnositel'nykh vkladov peremennykh v obshchuyu determinatsiyu v otsenennom na osnove metoda vypryamleniya iskazhennykh koeffitsientov regressionnom uravnenii // Vestnik SibGUTI. 2022. No. 1 (57). P. 89-96.

Bazilevskiy M.P. Metod vypryamleniya iskazhennykh iz-za mul'tikollinearnosti koeffitsientov v regressionnykh modelyakh // Informatika i ee primeneniya. 2021. Vol. 15. No. 2. P. 60-65.

Molnar C. Interpretable machine learning. – Lulu.com, 2020.

### Refbacks

- There are currently no refbacks.

Abava Кибербезопасность IT Congress 2024

ISSN: 2307-8162