Remember me
By 2030, the economic burden of chronic diseases is projected to reach approximately 343 trillion CNY (US $47 trillion) []. In China, the average direct medical expenditure for a case of lung cancer was 39,015 CNY (US $6041) in 2011, with an annual growth rate of 7.55% []. Meanwhile, Statista’s 2024 data reported that the average annual disposable income per capita in 2015 was just 21,966 CNY (US $3404)—showing how severe illnesses could push average-income families into poverty.
In light of these financial challenges, crowdfunding has emerged as a key charitable tool, enabling widespread, low-cost support for individuals facing medical expenses without expectation of reciprocation []. In China, platforms such as Shuidichou (also called Waterdrop) and Qingsongchou have been developed and supported more than 5 million patients from September 2014 to the end of 2021, raising more than 80 billion CNY (US $12.03 billion) []. However, medical crowdfunding projects on these platforms have low success rates: Only 9% of the projects reach their fundraising goals, while 70% achieve just 10% of their target []. This underscores the urgent need for research to improve crowdfunding success.
Previous medical crowdfunding research has largely examined success factors in isolation. These factors include cultural proximity [], social media outreach [], gender dynamics [], socioeconomic status [], linguistic style and structure [,], and donor-recipient relationships []. More recently, research has begun to explore the properties of information, extending to textual [-], visual [,,,,], and progress-related [,,,,] features of crowdfunding projects.
Most previous studies have focused on GoFundMe [,,,], which hosts a wide range of projects and primarily serves Western populations, leaving donor behavior in Asian contexts less understood. The majority of Chinese relevant research in China has been conducted using data from Qingsongchou [,,,] and Tencent Charity platform [,]. In contrast, Shuidichou, China’s largest medical crowdfunding platform, is dedicated exclusively to health care projects and remains understudied despite its relevance to Chinese users. While prior research has examined individual informational features, few studies have explored how different cues interact to shape outcomes. To address these gaps, this study applies Gibson’s affordance theory [], which emphasizes how environments provide actionable information that guides behavior. In the context of medical crowdfunding platforms, the platform serves as the environment, and crowdfunding projects act as information offered to potential donors. Overall, this study aims to address key gaps in prior research [-,,-], which has often focused on Western populations and overlooked the interplay among affordances.
Guided by this framework, our study explores: (1) how narrativity affordances (text length, patient demographics, sentiment) influence project success; (2) how visibility affordances (number and type of images) shape donor engagement; (3) how progress affordances (update frequency and sentiment) impact project outcomes; (4) how different types of information—narrativity, visibility, and progress affordances—shape success when controlling for other affordances within the same environment; and 5) how these affordances interact with demographic and contextual factors to shape crowdfunding success.
Building on prior studies, this research responds to Zhang et al [] by extending the analysis of narrativity affordances beyond project titles to include project descriptions and all textual content. It expands visibility affordances by drawing on Karahanna et al [] and Zhang et al [] to examine the number of images across projects. Inspired by Thies et al [], it investigates progress affordances—particularly update frequency—and incorporates update sentiment, following Zhang et al [], for a more nuanced content analysis. Additionally, building on findings by Zhou [], it includes demographic and project characteristics as control variables to assess how affordance effects vary across contexts. By synthesizing insights across modalities, this study constructs a comprehensive multimodal framework encompassing narrativity, visibility, and progress affordances.
Literature ReviewAffordance, as an ecological and psychological concept, has been developed and extended for decades. It was initially proposed by Gibson [], defining it as the information offered by the environment with references to the abilities of animals to act, mainly in the field of visual perception. Post-Gibson stream researchers [-] extend Gibson’s theories and summarize affordance as the property of the environment. Debates and discussions led to further development of Gibson’s theories. Norman [] was among the earliest to link affordance with human-computer interaction, laying the groundwork for a future extended definition of interaction-oriented affordance. He stresses that users are important because they can perceive and interpret affordances that are built into the design of computers and systems. Later, Jiao et al [] apply affordances to technologies. He denies and criticizes technological determinism and stresses that affordances are embedded in interactions among users, objects, and technologies. Through perceiving and understanding technologies, users can achieve different goals and outcomes. Heft [] emphasizes that users can take corresponding actions based on the provided information. Meanwhile, these actions influence information effectiveness.
The rapid rise of the internet and digital technologies has made social media platforms an integral part of daily life, supporting a wide range of activities such as self-promotion, marketing, entertainment, political engagement, and crowdfunding. In recent years, a growing body of academic research has explored the influence of platform affordances on user behavior. For instance, Karahanna et al [] proposed that individuals use social media to fulfill both psychological and emotional needs, with platform affordances—particularly IT-enabled ones—helping to meet these objectives. Several studies further highlight how platform affordances shape web-based behavior. Zhou [] finds that changes in social media affordances can causally affect communication among users. Theocharis et al [] examine how platform features and user interaction reshape political participation. Scharlach and Hallinan [] argue that IT-enabled affordances transform abstract social values into measurable actions on the web, guiding users’ decisions. In the context of medical crowdfunding, Choy and Schlagwein [] used the Earthship Kapita case to illustrate how IT affordances on crowdfunding platforms satisfy donors’ psychological needs and motivations, making web-based charity models more effective than traditional offline approaches. Jiao et al [] deliver a web-based survey and prove that IT affordances effectively influence donors’ perceptions and, therefore, donors’ donation decisions. Collectively, these studies emphasize the central role of affordances in shaping user behavior on the web.
Medical crowdfunding has emerged as a trending method for financing high-cost medical treatments. Severe diseases, exemplified by the far-reaching impact of the COVID-19 pandemic over the past few years, typically put significant economic pressures on patients and their families. Consequently, many resort to medical crowdfunding as a source of economic assistance. To better understand the mechanism of the practice and to optimize the chances of the success of crowdfunding projects, we frame medical crowdfunding as a multimodal information system facilitating the complex interaction between the three central stakeholders: (1) project initiator (organizer or patient), (2) the platform's technology and information infrastructure, and (3) prospective donors. We maintain that when information relevant to the project is properly presented and creates interaction between these actors, the chances of project success can be greatly improved.
[-,,,,,] displays related works and research data of medical crowdfunding platforms over the past several years. Based on the table, the primary studied medical crowdfunding platform includes GoFundMe, Tencent Charity, and Qingsongchou, while the primary methodologies used in related works are text analysis and linear regression. Independent and dependent variables are also acquired as seen in [-,,,,,]. Although various variables are chosen, related works only analyze the impact of the chosen variables on the project success in isolation.
Our study uses data from the Shuidichou platform, which serves a Chinese demographic, to examine both the individual effects of variables and the interaction. In contrast to prior research mentioned that has predominantly relied on data from GoFundMe—a platform primarily used in the United States—our approach expands the demographic scope by focusing on the Chinese context. As the largest crowdfunding platform in China, Shuidichou specializes in health care–related projects, offering a reliable and relevant data source for studying medical crowdfunding. While Qingsongchou is another Chinese platform with comparable functions, it presents a critical limitation: project organizers can alter fundraising goals during a project, potentially introducing bias and compromising data consistency. Therefore, Shuidichou provides a more targeted and stable foundation for our research. In addition to these, while each affordance may individually influence project success, their interactions must also be considered. Beyond studies using field data from platforms like GoFundMe that analyze affordances in isolation, previous experimental research—such as Jiao et al []—has used surveys to manipulate specific features and examine their effects. In contrast, our study uses objective field data from Shuidichou and considers affordances simultaneously, capturing how they function together in real-world settings. This integrated approach provides a more realistic and comprehensive understanding of crowdfunding dynamics and offers actionable insights on optimizing information presentation to enhance project effectiveness. In addition, clear explanations on how information presentation can be managed to enhance crowdfunding success are provided, along with examples.
Table 1. Medical crowdfunding affordance-related literature in the past 5 years.PlatformObs numberMethodFindingsSourceGoFundMe997Visual analytics approachSocial media plays a crucial role in boosting medical crowdfunding success.Ren et al []Qingsongchou1010Convolutional neural networksEmotional-related and credible-related photos increase the donation amount, while rational-related ones decrease it.Wang et al []Tencent Charity1891Linear regressionProject with health-related and disease-related keywords are more likely to succeedBa et al []Tencent Le Donation29,700Textual analysisPersuasive language significantly influences donor willingness and trust.Wu et al []Shuidichou50,000Linear regressionProjects with positive expressions achieve higher success rates.Zheng and Jiang []Qingsongchou754Linear regressionFunding goals and project duration can help to predict crowdfunding project performances.Chen et al []GoFundMe1830Univariate, multivariate analysisThe success rate of the project heavily depends on the cancer type and the donor’s social networks.Holler et al []GoFundMe92,753Textual analysisHard-to-treat, high-mortality cancers attract more donors. Projects with strong sentiment words and female beneficiaries are more likely to succeed.Zhang et al []GoFundMe243,795Topic modelImages featuring younger individuals, more people, and more smiles significantly improve the success of projects.Wang et al []Qinsongchou116,282Linear regressionAir pollution negatively impacts donations for critical illness, increasing negative emotions.Hua et al []Qingsongchou84,712Linear regressionKey factors such as patient age, disease type, geographical location, and donation target, influence the success of projects.Zhang et al []Table 2. Affordance indicators chosen by the works in Table 1.AuthorsIndependent variableDependent variableDrawing on affordance theory [], we conceptualize the crowdfunding platform as an environment in which textual, visual, and progress-related elements function as informational affordances. Similarly, signaling theory states the presence of information asymmetry between sender and receiver []. In web-based medical crowdfunding (), we have the sender as the crowdfunding project organizer, while the receivers are potential donors online. Senders use affordances to make information signals efficiently to influence receivers, and receivers make informed decisions based on the signals. All interactions among receivers and senders and the information presented on the platform exert a significant influence on donor behavior, ultimately shaping the success rate of crowdfunding projects.
Figure 1. Interaction between project organizers or patients, potential donors, and the platform. To quantitatively evaluate a project’s success, the most frequently applied metric is the project’s success rate, although its definition is not uniform across earlier studies. For example, Ba et al [] use a variable termed Goal_Pro, where the percentage of raised funds compared with the target donation is used. Likewise, Zhang et al [] follow the same benchmark. By way of contrast, Zheng and Jiang [] gauge success by measuring the number of donations received and the donation amount. A lesser-used measure is binary success—whether a project is a success. For example, Parhankangas and Renko [] create a dummy variable where 1 represents a goal achieved and 0 represents failure. Both continuous and binary measures are used by our study for the dependent variables, providing a richer assessment of project results. Based on these theoretical underpinnings and earlier results ( and ), the following hypotheses are proposed.
We define narrativity affordance as the action possibilities offered by textual elements written by project organizers. On Shuidichou, these include the project title, project description, fundraising purpose, donation plans, and surplus fund use (). To capture donor attention, patients and platform managers must craft descriptions using effective language. Multiple indicators, including text length and sentiment analysis, are selected to evaluate narrativity affordance.
The text lengths of titles and project descriptions have been widely studied, yet findings remain inconsistent. Hou et al [] and Chen et al [] claim that a longer text length, especially the length of the story, can attract more attention. However, Zhang et al [] conclude that there is an inverted U-shaped relationship between story length and the success of a project, meaning that a project with a medium-length description is more likely to succeed. Similarly, Ren et al [] conclude that the optimal length of a successful project title should be 6 to 11 words. Zhang et al [] claim that statistically, the length of the title is not significant, and a shorter project description positively influences the success of the project. We develop our first hypothesis as follows.
Text length has an inverted U-shaped relationship with the success of a project.Studies also discussed whether project titles or project descriptions should contain disease type, gender, and age in order to improve the success of the project. Chen et al [], Hou et al [], and Peng et al [] all conclude that projects for younger patients, especially children, are more likely to succeed. Zhang et al [] confirm that if the project title includes disease type, the success rate of the project increases. No past research has directly proved that gender information included in the title and description can impact the success rate of a project. However, Holler et al [] conclude that female patients are more willing to initiate crowdfunding projects. We develop our second hypothesis as follows.
Disease type, gender, and age in the project title and the project description positively influence the success of a project.
Figure 2. Screenshots of narrativity affordances. The sentiments in the title and the project description also influence the success of a project. Researchers apply various methods to capture sentiments. Kim et al [] designed questionnaires and divided project sentiments into 3 types: responsibility, guilt, and urgency. In the experiment of Wang et al [], sentiments are measured using scales that include pride and guilt. Although questionnaires and experiments can measure sentiments more precisely, their limitation lies in the need for fully qualified participants; otherwise, accuracy may be compromised. Current techniques can be divided into 2 groups: Lexicon-based and machine learning methods. Agarwal et al [] construct a domain-specific ontology with ConceptNet and WordNet, extract wordings with natural language processing (NLP), and convert them into a contextual polarity lexicon, classifying sentiments into positive, negative, and neutral ones. Similar lexicon-based methods are used by multiple research works [-]. Machine learning techniques are commonly used by extracting and crawling keywords, followed by classification. With such techniques, it is possible to analyze a relatively larger volume of texts and quantify the sentiments in them [-].
Previous studies report various findings on sentiments in titles and project descriptions. Zhang et al [] conclude that the negative sentiments in project titles decrease the success rate, but if these words are in project descriptions, the success rate increases. Hou et al [] and Zheng and Jiang [] confirm that if the project description narratives are more positive, the project tends to succeed. We propose a hypothesis as follows.
Positive or negative sentiments of the project title and project descriptions significantly influence the success of a project.We define visibility affordance () as the action possibilities through design and graphical elements, specifically referring to the number of photos in different sections (main page and progress updates). Studies have proven that photos and videos on social media are effective tools for establishing web-based identification and attracting the attention of online users [,]. Similarly, posting photos for medical crowdfunding projects can provide potential donors with more direct information, and photo types, such as the current status of patients and ID verification, can influence donor motivation and donor trust []. Wang et al [] list the influence of different types of photos on the success of a project. They point out that photos related to the current health status and treatment status positively influence the success of a project, while those presenting treatment expenses negatively influence the success of a project. Thies et al [] and Wu et al [] conclude that the number of photos included positively influences the success of a project. Therefore, our hypothesis is as follows.
Figure 3. Screenshots of visibility affordances. The number of photos included in a project positively influences the success of a project.We define progress visibility () as the action possibilities offered by progress updates, including elements such as patient treatments, donation cash flows, etc. The progress updates of a project can be crucial for the following reasons. First, regularly posting updates establishes the credibility of the project organizer [], because this shows that the project is transparent and that donors can track their donations. Second, platform managers can structure textual descriptions and use words that convey positive or negative sentiments to evoke donor empathy. Some prior academic research has already highlighted the impact of progress updates. Parhankangas and Renko [] claim that a higher frequency of progress updates improves donor engagement, thereby enhancing the project’s success. Ren et al [] and Wu and Peng [] have similar conclusions. Zhang et al [] conclude that a high frequency of progress updates with negative sentiment words can significantly increase the success rate of a project. Wu and Peng [] propose that sentiments in progress updates significantly influence donation amounts. Previous studies also suggest an inverted U-shaped relationship between progress update frequencies and project success. Thies et al [] and Wu et al [] mention that when the frequency is too high, donors can be negatively influenced by information overload. We develop 2 hypotheses as follows.
The frequency of progress updates has an inverted U-shaped relationship with the success of a project.
Figure 4. Screenshots of progress affordances. This study did not require ethics board review because it did not involve direct interaction with human participants nor the collection of any private or identifiable information. All analyses were conducted on aggregated, anonymized data from public sources. This rationale is consistent with prevailing ethical research standards and common institutional practices for studies of this nature.
OverviewAll data were extracted from disease-related crowdfunding projects on the Shuidichou platform. This study did not involve any surveys or experimental interventions; thus, no RCT registration number is applicable. Launched in July 2016, Shuidichou is China’s leading platform for medical crowdfunding. By the end of 2020, the Shuidichou platform had facilitated over 37 billion RMB (approximately US $5.1 billion) in donations from more than 340 million contributors across 1.7 million medical aid projects, according to its publicly reported data.
According to company profile summaries on Tracxn.com, while GoFundMe remains the leading crowdfunding platform [], Shuidichou specializes in health care, making it China’s top platform and the world’s third-largest. Backed by US $251 million from major investors like Tencent and IDG Capital, its scale and user base offer a strong basis for analyzing information affordances in crowdfunding success.
On Shuidichou, project pages follow a standardized format. Information—such as the project title, target amount, funds raised, timeline, beneficiary, initiating organization, project leader, and donation recipient—is shown on the right side. On the left, donors can navigate three sections: (1) Project Details, which outlines the patient’s story, fundraising purpose, fund allocation, execution plan, and surplus handling; (2) Project Progress, which provides updates on the patient’s condition and fund use; and (3) Donation Status, which tracks contributions from donors.
To ensure representative results, this study uses a sample of 1261 crowdfunding projects initiated between May 11, 2018, and September 5, 2024. Due to platform restrictions requiring additional authorization for long-expired projects, we extracted data from all publicly available projects as of November 11, 2024, without applying additional selection criteria. As such, we consider the dataset broadly representative of various disease types, while acknowledging that its demographic scope is limited to the Chinese population and that some projects may have been overlooked due to limitations in platform visibility. Among these projects, 1228 out of 1261 (97.38%) have already closed. Given that this study aims to understand how the platform's information management capabilities influence whether a project reaches its funding goal, and recognizing that projects may meet their objectives before the intended end dates, we chose not to exclude any projects based on their completion status.
Data ExtractionThe Shuidichou platform restricts the direct extraction of comprehensive project data. To address this, research assistants manually collected project links to include all available projects for individuals with various diseases. Software engineers then used Python to scrape donor-visible data. Since patient gender, age, and name are not explicitly displayed in a dedicated section, these variables were manually coded.
After removing duplicates, we conducted a comprehensive textual analysis to examine how affordances influence donation behavior across project titles, project descriptions, progress updates, fund purposes, fund use plans, and surplus fund use. All text elements were evaluated for length and sentiment using SnowNLP. Titles and project descriptions were the primary focus, as they conveyed crucial information to donors. We measured text length excluding spaces and punctuation to assess information density and generated dummy variables for age, gender, and disease presence using Python-based text-matching methods.
For sentiment analysis, we preprocessed text using the Jieba Chinese segmentation tool and generated sentiment scores with SnowNLP, which assigns values from 0 (negative) to 1 (positive). SnowNLP is a widely used tool for Chinese sentiment analysis, particularly effective for processing user-generated content such as social media and crowdfunding texts. Prior studies have successfully applied SnowNLP for sentiment analysis in Chinese platforms, demonstrating its suitability for this context []. We further analyzed the frequency of positive, negative, and degree words using the Hownet Chinese dictionary [].
Variables and ModelsDependent VariablesThe success of crowdfunding projects is measured in two ways: (1) success ratio: the ratio of the donations received to the funding goal. (2) success indicator: a binary indicator of whether the project met or exceeded its goal. Only 39 projects have fully reached their funding targets. Given the limited number of successful cases, we conducted bootstrap robustness tests for each model using success_ind as the dependent variable. Only models that passed the robustness test—indicating low bias and stable results—were considered.
Independent VariablesThe main independent variable in this study is information affordance, comprising three components: (1) narrativity affordance, which examines how textual elements influence donation behavior through factors such as text length, word frequency, and sentiment scores; (2) visibility affordance, which explores the role of visual information, particularly the number of images in project descriptions and progress updates, in shaping donor engagement; and (3) progress affordance, which analyzes how the frequency and content of patient progress update impact donation patterns over time.
Control VariablesThe control variables in this study are patient age, patient gender, and an indicator of whether the beneficiary is an individual or a group of recipients.
The multivariable models assess how narrativity, visibility, and progress affordances influence crowdfunding success. Each model is tested with both the success indicator and success ratio using regression analysis and U-tests. Given that each affordance includes distinct elements—such as textual length, which may be reflected in title length, story length, or other components—that may influence donor decisions differently, we analyze them separately to test for differential effects. Models 1-5 assess the individual impact of specific affordance elements on the success ratio, while model 6 takes a holistic approach to examine how narrativity, visibility, and progress affordances collectively influence success.
Models 1 to 3 focus on narrativity affordance, leveraging the textual richness of titles and project descriptions. Model 1 examines titles, model 2 analyzes project stories, and model 3 combines both, along with additional elements such as purpose statements, plans, and surplus fund use, to assess the impact of narrative elements on donation behavior. Model 4 captures visibility affordance by measuring the effect of image counts in project descriptions and progress updates. Model 5 addresses progress affordance by incorporating patient progress updates’ frequency, length, and sentiment to explore how updates shape donation behavior.
Model 6 integrates the 3 affordances into a comprehensive framework. To assess the effect of each affordance type while controlling for the others, we aggregated related variables within each category. Narrative length is calculated by summing all relevant text fields. Demographic information—patient age, gender, and disease type—is combined into a composite index, with higher values indicating more demographic content. Wording valence is derived by subtracting total negative word counts from total positive word counts. Additionally, degree words and their intensity scores from both the title and project descriptions are summed to capture overall linguistic intensity.
Only 39 out of 1261 projects (3%) successfully reached their funding goals. On average, the projects achieved only 18% of their target. To validate our success measures, we examined their correlations with key outcomes. Both the number of donations (R=0.5512 and 0.2359) and the amount received (R=0.5455 and 0.2167) show strong, significant correlations with success ratio and success indicator (all P<.001). Descriptive statistics are provided in Tables S2 and S3 in .
Robustness TestTo ensure the statistical validity of our results, we conducted robustness tests for both the success ratio and success indicator models. Given the small number of fully successful projects, these tests help confirm that our findings are stable and not driven by random variation. Specifically, we used nonparametric bootstrapping with 1000 replications for models 1 to 6 using both dependent variables.
For the success ratio models, bootstrap analyses demonstrated generally stable coefficient estimates across models 1 to 6, with low bias (< ±0.05) and moderate standard errors (<0.10). However, in model 2, the sentiment score calculated using the SnowNLP algorithm exhibited substantially inflated standard errors (up to 30) and high bias, indicating instability for this predictor. Similarly, in model 3, the indicator for whether the project beneficiary is an individual showed large standard errors (up to 1464.41) and considerable bias, likely due to model complexity and multicollinearity among narrative-related features. These unstable predictors were excluded from interpretation, while the remaining estimates across models 1 to 6 demonstrate solid robustness in explaining the success ratio.
In the success indicator models, bootstrapping with 1000 resamples for models 1 to 3 (focused on narrative affordances) revealed substantial statistical instability. Several predictors showed significant bias (eg, exceeding ±1.0 in model 1 and reaching up to ±1.78×1013 in model 3) and substantial standard errors (up to 4.65×1014). Consequently, estimates from these models are excluded from further interpretation. In contrast, bootstrap analyses for models 4 to 6, which examine visual, progress, and combined affordances, produced substantially more stable results. Most key predictors in these models exhibited minimal bias (< ±0.05) and moderate standard errors (<0.10), supporting the reliability of these estimates. Therefore, models 4 to 6 are retained for interpretation, given their demonstrated stability.
Collectively, these analyses demonstrate that the majority of models in this study produce stable and reliable estimates, with specific unstable predictors identified and excluded from interpretation. To further assess the robustness of the success ratio models, we conducted a sensitivity analysis by re-estimating the models after excluding projects that fully met or exceeded their fundraising goals. This approach tested whether these extreme cases disproportionately influenced the results. The findings remained consistent, with similar coefficient magnitudes and significance levels, and only minor fluctuations were observed. All results based on the filtered dataset are reported for transparency (Tables S4-S6 in ). Together, the bootstrap and sensitivity analyses provide a comprehensive assessment of model reliability, while acknowledged limitations ensure cautious, evidence-based interpretation.
Regression ResultsOverall, we present 6 models using the success ratio (-) and success indicator ( and ). U-shaped relationships are visualized in -.
Table 3. Regression results for success ratio (models 1-3). This table presents regression results for narrative affordance models (models 1-3) with success ratio as the dependent variable. Model (a) shows main effects; model (b) adds U-tests for significant predictors.aVariable was not available and therefore not included in the model.
bN/A: not applicable, but included in the model under every table.
Table 4. Regression results for success ratio (models 4 and 5). presents regression results for visual and progress affordance models (models 4 and 5) with success ratio as the dependent variable. Model (a) shows main effects; model (b) adds U-tests for significant predictors.
Comments (0)