EN
The paper presents arguments that advocate for application of the multi-group confirmatory factor analysis as a tool for constructing sentiment indicators in business surveys. Reliable measurement and comparisons of the sentiment mean between periods require measurement invariance on its three basic levels-configural, metric and scalar invariance. It is hypothesized that only sets of questions that are internally coherent can serve as a group of proxies for business sentiment indicator. An attempt to construct two different sentiment indicators for manufacturing industry is performed. The results show that only for the set of coherent proxies it is possible to estimate model with measurement invariance.