All time references are in CEST
Multilevel approaches to survey data analysis |
|
| Session Organiser | Dr Dalila Failli (University of Florence) |
| Time | Wednesday 19 July, 09:00 - 10:30 |
| Room | U6-11 |
Data with hierarchically structured units, also known as multilevel data, are often encountered in survey research. In this context, a multilevel approach offers a very useful methodological tool to properly account for the hierarchical structure of the data. Indeed, this type of approach allows not only to deal with unobserved sources of heterogeneity at multiple levels, but also to take into account the possible correlation between the responses provided by units belonging to the same hierarchical level. This session is specifically focused on multilevel approaches to survey data analysis. In this regard, the first contribution of the session extends a model-based clustering method in a multilevel context for the analysis of network data, with the aim of analyzing the gray digital divide among different European countries. A different use of the multilevel approach is that of the second contribution, in which a multilevel regression with post-stratification is performed to model individual survey responses based on demographic and geographic predictors in order to estimate voting choice in the 2023 Estado de México governor election. Finally, the third contribution of the session exploits the hierarchical nature of organizational survey data to explore the antecedents of organizational-level engagement in a setting in which employees are nested within service sector organizations in India.
Keywords: multilevel data, model-based clustering method,
Ms Pavithra Ganesh (Indian Institute of Technology Kharagpur) - Presenting Author
Professor Kailash B L Srivastava (Indian Institute of Technology Kharagpur)
The application of multilevel analysis on organizational survey data has been an area of study and debate in the fields of psychology and management. This is due to the nested or hierarchical nature of organizational data. Since employees are nested within organizations, the independence of their responses has been called into question. Similarly, measurement of composite constructs also ignores this hierarchy since researchers usually use the average of scores to measure the variable at the organizational level. It is only recently that multilevel analysis has been applied to organizational data in order to mitigate these methodological concerns. The present study aims to demonstrate the use of multilevel analysis on survey data collected from 704 employees in 147 service sector organizations in India. Robust multilevel structural equation modelling with Mplus software was conducted to test the impact of HR practices and organizational culture on the emergence of collective organizational engagement. Cross-level invariance was also tested for all the study variables at within and between levels. Collective organizational engagement is a multilevel construct which has been defined as the shared perception that employees are physically, cognitively and emotionally engaged in their work. Results showed that HR practices and a positive organizational culture were antecedents of collective organizational engagement at the individual and organizational level. This study is one of the first to explore the antecedents of organizational-level engagement using multilevel analysis, thereby adding to engagement literature. The study also has methodological implications for researchers and practitioners, since it exemplifies the advantages of understanding the hierarchical nature of organizational survey data.
Dr Dalila Failli (University of Florence) - Presenting Author
Professor Bruno Arpino (University of Padua)
Professor Maria Francesca Marino (University of Florence)
The grey digital divide is the gap between older adults and younger individuals in accessing the latest information technologies. Although it has narrowed over the past decades, it still remains substantial, and older people show a lower propensity to have a broadband connection, use the Internet, and adopt new technologies. Motivated by the analysis of the grey digital divide, we build a bipartite network concerning the presence of various digital skills in individuals from different European countries. Bipartite networks are particularly useful for representing relationships between two disjoint sets of nodes, formally called sending and receiving nodes. The goal is to perform a clustering of individuals (sending nodes) based on their digital skills (receiving nodes). In this regard, we extend the Mixture of Latent Trait Analyzers (MLTA) model in a multilevel setting for the analysis of bipartite networks, also accounting for nodal attributes in order to analyze how socio-economic and demographic characteristics, as well as intergenerational ties, influence the individual digitalization level. The proposed model is not only able to cluster sending nodes, as in the latent class model, but also to capture the latent variability of network connections within each cluster, as in the latent trait framework. In addition to the analysis of data from the European Social Survey (round 10), a simulation study is also conducted to test the performance of the model in terms of parameters and clustering recovery.
Mr Yamil Nares (Defoe)
Mr Alfredo Chaparro (Defoe)
Miss Alexa Martínez (Defoe) - Presenting Author
Multilevel regression with post stratification (popularly known as “MR. P”) is a technique that uses multilevel regression to model individual survey responses based on demographic and geographic predictors, “post-stratifying” estimates with respect to actual demographic and geographic proportions.
In this study, we explored MRP as a tool to estimate vote choice in the 2023 Estado de México governor election. These elections have historically been a preliminary stage of the elections for president of Mexico 2024, it is also the state where the largest proportion of people on the nominal list is concentrated.
The objective of the study is to apply the technique in the local Mexican electoral case to provide a comparison of the estimators against the official results at state, district and municipal levels. The surveys will be obtained through the social network Facebook and other sources, addressed to residents of the entity, the main goal is to capture trends in vote intention during the 2023 electoral contest. The surveys will be applied from January to June in 2023.
Two times in a month, a multi-party categorical Bayesian multilevel regression model will be applied to the electoral preference considering demographic variables and geographic information as a level variable. With the results, a probability prediction imputation will be made to the different combinations of population characteristics used in the model, and the corresponding post-stratification will be applied.
We will use actual state level results to assess estimates resulting from MRP. The results are also analyzed by the main demographic groups.