Exploring Innovation in Nonprofit Organizations: Topic Modeling of Articles of Incorporation and Multilevel Analysis

In times of political and economic turbulence across countries, nonprofit organizations, including human service organizations (HSOs), are facing growing challenges that demand innovative managerial practice. Traditionally, research on the management of HSOs has relied on surveys and interviews. However, with increasing attention to the use of AI and computational methods in research, there is growing interest in incorporating organizational-level textual data, such as administrative documents to better understand managerial dynamics. Despite this need, few studies have combined survey data with organizational textual data.

To fill this gap, this study examines factors associated with organizational innovation in nonprofit organizations by using multiple data sources, including surveys and textual analysis. In particular, it focuses on how mission-related text in nonprofit organizations’ articles of incorporation are associated with perceived organizational innovativeness. Research question asks: What types of organizations, based on their goals and activities, are more prepared to innovate programs and implement new ideas?

This study adopts a mixed-methods design, combining text analysis of organizational documents (article of incorporation) with multilevel modeling using survey data. Data collection occurred in two stages. First, survey data were collected from 565 employees across 232 randomly selected NPOs in South Korea. Survey questionnaires included employees’ perceptions on environmental turbulence, innovativeness, mission clarity, along with demographic characteristics (age, gender, positions). In parallel, articles of incorporation were gathered from the same organizations. From these documents, mission-related texts reflecting each organization’s direction and identity were extracted. Texts were analyzed using standard natural language processing (NLP) techniques, including tokenization, stopword removal, and stemming using Textom for initial preprocessing and Python for topic modeling. We then applied Latent Dirichlet Allocation (LDA) topic modeling to identify latent themes across the corpus. A four-topic solution was selected based on coherence and interpretability. The four topics and their proportions were: Topic1(28.5%) Protection and care for the socially disadvantaged; Topic2 (19.8%) Education and socio-cultural research; Topic3 (15.6%) Citizen participation in justice; Topic4 (36.1%) Welfare services and community practice.

Multilevel modeling results organizational mission types, as stated in articles of incorporation, were significant factors associated with perceived innovativeness. The intraclass correlation coefficient (ICC) on four different types of innovation were ranges from .19 to .23, indicating that approximately 19% to 23% of the variances in innovation was attributable to differences between organizations. Compared with citizen participation organizations, NPOs focusing on education and research activities (Topic2) and protection and care for the socially disadvantaged (Topic1) were more likely to show conceptual innovation. However, no significant difference were found for program development or implementation innovation. Organizations in Topic 2 and Topic 4 were more likely to show strategic adaptive innovation (b=12, p<.001; b=11, p<.05). Perceptions of shared missions and organizational turbulence were also associated with perceived innovativeness at the individual levels.

This study suggests mission types influence on the level of innovativeness. From the social work research perspective, it demonstrates the utility of integrating textual data and survey responses to further managerial research in human service organizations.