Access to this full-text is provided by IGI Global Scientific Publishing.
Content available from Journal of Global Information Management
This content is subject to copyright. Terms and conditions apply.
DOI: 10.4018/JGIM.344044
Volume 32 • Issue 1
This article published as an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/) which permits unrestricted use, distribution, and production in any medium,
provided the author of the original work and original publication source are properly credited.
*Corresponding Author
1
Dongmei Han, Shanghai University of Finance and Economics, Shanghai, China
Lifeng He, Qingdao University, China
Wenfei Zhao, Shanghai University of Finance and Economics, China
Xiaohang Zhou, Qingdao City University, China*
https://orcid.org/0000-0001-8846-1845
Given the unwieldy glut of information in online health question-answer (Q&A) service, it is essential
to understand what constitutes helpful answers in the medical domain. Despite the fact that studies
have examined the impacts of answer content factors on answer helpfulness, there are two gaps that
need further analysis. First, the empirical results of the existing relevant studies on the effect of answer
emotion are inconsistent. Second, prior studies only have examined the independent impacts of answer
content factors and question content cues on answer helpfulness. To fill this gap, a research model
reflecting the impacts of emotional content and question-answer congruence on answer helpfulness
was developed and empirically examined. Our empirical analyses confirm that emotional content
and answer helpfulness are related to one another in the form of an inverted U-shape and indicate
that two types of question-answer congruence (emotional intensity congruence and linguistic style
matching) positively affect answer helpfulness. Theoretical and practical implications are discussed.
Answer Helpfulness, Emotional Content, Linguistic Style Matching, Online Health Q&A Community, Question-
Answer Congruence
Online health platforms are flourishing and have fundamentally altered the manner in which the
general public seeks health information and advice (Harris Interactive, 2011; Tan & Goonawardene,
2017; Huang & Wang, 2022). Two-thirds (67%) of participants in the American Health Information
National Trends Survey from 2008 to 2017 reported turning to the internet first for health information
(Finney et al., 2019). Similarly, 69.3% of Canadians indicated that they used the internet to search for
Volume 32 • Issue 1
2
health-related information in 2022 (Statistics Canada, 2023). To address this urgent and substantial
demand from the public, numerous platforms offer services to facilitate access to health-related
information, such as the question-and-answer (Q&A) function module, through which individuals
may seek health advice by posing questions and assist others by offering answers (Silberg et al., 1997;
Peng et al., 2020). Online health communities have recently been increasingly supporting mental
health amid a growing global burden of mental disorders and limited resources (Yan et al., 2022).
Stigma around mental health can deter or delay individuals from seeking professional help, worsening
outcomes (Xiang et al., 2012). Consequently, the practice of accessing mental-health information
online has emerged as an effective solution (Zhou et al., 2019), and online mental-health communities,
which link information providers with those in need, have gained popularity accordingly (Xiang et
al., 2012; Yan et al., 2022).
Although the general public frequently trusts online health-related information, it is not always
reliable (Silberg et al., 1997; Harris Interactive, 2011; Peng et al., 2020). In fact, many online health
communities permit individuals lacking medical expertise to provide advice owing to their open
nature (Jin et al., 2016). Studies also indicate that the majority of health information acquired online
is not authored by professionals in the medical field (Toma & D’Angelo, 2015). Given the significant
impact of health information on individuals’ health decisions and the variable quality of online health-
related information (Peng et al., 2020), it is imperative for patients, health communities, and society
at large to deepen their understanding of individuals’ value judgments (i.e., perceived helpfulness
value) regarding online health information (Toma & D’Angelo, 2015; Jin et al., 2016; Peng et al.,
2020; Zhang et al., 2020).
To determine the perceived value of information (answers) provided by repliers (individuals
who respond to questions and offer advice) (Jin et al., 2016), numerous online Q&A communities
have established a helpfulness voting system, where readers (i.e., individuals who browse, read, and
evaluate answers in online health Q&A communities) can vote on helpful answers and the vote counts
(i.e., the number of helpfulness votes) are visible to all. This voting system aids in identifying more
helpful answers and lessening information overload for readers (Jones et al., 2004; Peng et al., 2020).
Consequently, understanding the factors influencing the perceived helpfulness of answers enables
future readers to more readily access helpful information and guides repliers in crafting more helpful
answers (Peng et al., 2020). Furthermore, analyzing readers’ voting behaviors regarding helpfulness in
online health communities not only provides insights into the design and improvement of guidelines
for helpful answers but also facilitates the development of recommendation and prediction systems and
boosts the effectiveness of auxiliary detection tools within these communities (Jin et al., 2016; Zhang
et al., 2019; Liu et al., 2020; Peng et al., 2020). More broadly, identifying these factors influencing
helpfulness is pivotal in enhancing the provision of high-quality public health information and in
aiding the decision-making process in the health-care domain (Harris Interactive, 2011; Toma &
D’Angelo, 2015).
Recognizing the practical significance of identifying answer helpfulness, an increasing number
of scholars have embarked on research in this area (Peng et al., 2020). Prior studies have identified
various factors that impact the perceived helpfulness value of answers within online Q&A communities
(Jin et al., 2016; Zhang et al., 2019; Peng et al., 2020). Employing knowledge-adoption theories (i.e.,
the elaboration likelihood model and the heuristic systematic model), this body of work focused
primarily on answer-content characteristics (i.e., answer length, emotion, and readability) (Jin et
al., 2016; Liu et al., 2020; Zhang et al., 2020) and source characteristics (such as reputation and
expertise of repliers) (Jin et al., 2016; Zhang et al., 2020). Studies have also examined how question
characteristics, serving as critical contextual cues, influence readers’ value assessment of answers
(Chung & Park, 2013; Liu et al., 2020).
The current investigation builds upon previous work in two significant ways. First, this study
emphasizes the role of emotional content—a key characteristic of answers—in shaping readers’
perceptions of answer helpfulness. Emotion refers to a person’s subjective sentiment and feelings
Volume 32 • Issue 1
3
stemming from the evaluation of goals pertinent to an individual’s well-being (Wang et al., 2019). In
online health Q&A communities, the role of emotional expression in answers appears increasingly
significant. This is attributed to the fact that, beyond the pursuit of information to mitigate uncertainty,
individuals seeking health-related information also necessitate emotional support (manifested in
forms like comfort, empathy, and understanding) (Kazmer et al., 2014; Jin et al., 2016; Cronin et al.,
2018; Zhang et al., 2020).
Previous studies have investigated how emotional content influences readers’ perceived
helpfulness of answers in online Q&A communities (Jin et al., 2016; Peng et al., 2020; Zhang et
al., 2020). However, these findings are inconsistent and incomplete. On the one hand, emotional
content can increase perceived answer helpfulness by providing emotional support (Jin et al., 2016)
and facilitating information processing (Silberg et al., 1997; Jin et al., 2016). On the other hand,
emotional content can decrease the objectivity (Liu et al., 2020) and perceived expertise of answers
(Zhang et al., 2019), potentially detracting from their perceived helpfulness. These inconsistencies
indicate that the impact of emotional intensity on answer helpfulness is nuanced and complex and that
relying solely on a single mechanism is insufficient to explain the influence of emotional intensity
on the helpfulness of answers.
Subsequent research into the effect of emotional intensity (the percentage of expressed emotions
in the content) on perceived value of information reveals that emotional intensity has both bright sides
(such as emotional support) and dark ones (such as decreased objectivity) (Toma & D’Angelo, 2015;
Yin et al., 2017). At low levels of emotional intensity, an increase in emotional expression tends to
correlate with greater perceived helpfulness of information (Yin et al., 2017; Li & Huang, 2020).
Conversely, at high levels of emotional intensity, a surge in emotional expression may correlate with
a decline in the perceived helpfulness of the information (Yin et al., 2017). Given the multifaceted
nature of emotional intensity’s influence, it is imperative to further investigate its nonlinear effects
on the perceived helpfulness of answers.
Second, although researchers have made significant progress in understanding the impacts of
answer characteristics and question factors on a reader’s helpfulness value assessment of an answer
(Silberg et al., 1997; Toma & D’Angelo, 2015; Jin et al., 2016; Zhang et al., 2020), these two elements
are not distinct and separate. Indeed, readers do not evaluate answer (or question) characteristics in
isolation but always consider question-and-answer characteristics in tandem (Chung & Park, 2013;
Peng et al., 2020). For example, existing research has explored the role of question and answer content
relevance (topic relevance) on the helpfulness of answers (Liu et al., 2020; Zhang et al., 2020).
However, there is a notable gap in the literature regarding the relationship between question–answer
congruence in linguistic attributes (linguistic features) and the helpfulness of answers. The effects of
congruence in linguistic attributes versus content relevance differ significantly (Yang et al., 2021).
Content relevance mainly indicates the degree of matching in terms of topics and expertise (Liu et
al., 2020; Yang et al., 2021). Linguistic attribute congruence significantly influences readability
and diagnosability for readers (Zhou et al., 2020; Yang et al., 2021). Many studies underscore the
importance of congruence in linguistic attributes for facilitating processing, enhancing fluency,
and fostering a common social identity (Zhou et al., 2020; Yang et al., 2021; Zhang et al., 2021).
Furthermore, grasping the nuances of question–answer congruence in linguistic attributes and its
effects on answer helpfulness is crucial for designing recommendation and prediction systems that
leverage linguistic features. Thus, it is essential to understand answer helpfulness by considering
question–answer congruence in language attributes.
To fill in this gap, our research addresses two questions: (1) What is the effect of emotional
content on answer helpfulness in the online health Q&A community? (2) How does the question–
answer congruence affect answer helpfulness in the online health Q&A community? Specifically,
our research proposes that emotional-content intensity is related to answer helpfulness in an inverted
U-shape. Further, we propose that two types of question–answer congruence—emotional-intensity
Volume 32 • Issue 1
4
congruence and linguistic style matching (the congruence in usage of function words)—influence
the reader’s perceived helpfulness of answers.
To address these two research questions, we collected data from a leading online health Q&A
community in China, and then we coded the variables using Linguistic Inquiry and Word Count (LIWC,
Pennebaker et al., 2001), a tool that has been used in many fields, including information systems
literature (Huang et al., 2017; Zhang et al., 2021). We applied a replier-level fixed-effects negative
binomial regression model in the main test and addressed several possible endogeneity problems in
the robustness check. Empirical results showed that emotional intensity is related to perceived answer
helpfulness in an inverted U-shaped relationship. In addition, we found that emotional-intensity
congruence and linguistic style matching have positive effects on answer helpfulness and that these
two types of question–answer congruence can jointly affect answer helpfulness.
User-generated content (UGC), such as reviews on e-commerce sites (Cheng & Yang, 2022; Pham et al.,
2023) and answers on Q&A sites (Jin et al., 2016; Zhang et al., 2020), constitutes a critical component
of online communities. To locate and highlight helpful UGC and reduce confusion, many communities
have established helpfulness-assessment systems. A large number of studies have been dedicated to
understanding what constitutes perceived UGC helpfulness. For online reviews, prior research has
indicated that factors influencing review helpfulness primarily include content characteristics such
as review length, expressed emotion, and readability (Mudambi & Schuff, 2010; Eslami et al., 2018;
Filieri et al., 2018). Additionally, scholars have also given attention to the specific characteristics of
the reviewer, including anonymity and reputation (Cheung et al., 2012; Hama et al., 2019).
In recent years, owing to the development of online Q&A communities, scholars have intensified
their examination of the perceived helpfulness of answers. Some studies have examined replier
characteristics, including reputation score (Chen et al., 2010), expertise (Edwards, 2008), and
experience (Shah & Pomerantz, 2010). Additionally, question-level factors such as question length
(Shah & Pomerantz, 2010), attribute category (Harper et al., 2008), and reward size (Jeon et al.,
2010) have also been examined to determine how they affect the perceived helpfulness of answers.
Subsequently, a greater portion of the literature has concentrated on the characteristics of answer
content. For example, answer content characteristics, such as answer length (Silberg et al., 1997;
Liu et al., 2020; Zhang et al., 2020), politeness (Lee et al., 2019), metaphor (Zhang et al., 2019), and
readability (Liu et al., 2020), have been subject to examination. Table 8 in the Appendix summarizes
the examined influence factors of the perceived value of answers in online Q&A communities.
Although emotion, as a significant component of answer content characteristics, has been
extensively researched in the context of answer helpfulness (especially health-related answer
helpfulness), researchers’ findings were mixed (Jin et al., 2016; Yin et al., 2017; Li & Huang,
2020; Liu et al., 2020; Zhang et al., 2020). On the one hand, Jin et al. (2016) showed that higher
emotional intensity can provide more emotional support for readers, thereby potentially increasing
the perceived helpfulness of the health-related answer. In addition, emotional expression can increase
the perceived helpfulness by facilitating readers’ information processing (Li & Huang, 2020). On
the other hand, the literature also indicates that emotional content could detract from the objectivity
and perceived expertise of health-related answers (Zhang et al., 2020). These inconsistent findings
suggest that emotional content may influence the reader’s perception of answer helpfulness in a
more sophisticated manner. Certain prior studies have adopted a reductive approach by assuming
that the relationship between emotional content and perceived value is linear. However, as suggested
by Yin et al. (2017), content with low and high emotional intensity may elicit divergent perceptions
among readers, potentially giving rise to an inverted U-shaped relationship rather than a linear one.
To elucidate the aforementioned inconsistent findings, our study will investigate the potential for
Volume 32 • Issue 1
5
an inverted U-shaped relationship between emotional content and answer helpfulness within online
health Q&A communities.
Despite the extensive research into question content and answer content, the interaction between
them has not been thoroughly explored. Readers initially encounter the question content in the
thread before evaluating the answer content. Therefore, the perceived helpfulness of answers may
be significantly shaped by their interaction with the corresponding questions. Several studies have
suggested that the impact of focal review content on helpfulness depends on contextual cues, including
previous reviews and management-response text features (Yin et al., 2016; Yang et al., 2021). Given
the critical nature of these interactions and the distinct context of online health Q&A communities,
this study posits that the influence of an answer’s emotional content on its perceived helpfulness is
potentially modulated by the emotional intensity of the associated question. To further explain the
inconsistent findings about the relationship between the answer’s emotional content and the answer’s
helpfulness, from the perspective of question–answer interaction, we argue that the emotional-intensity
congruence between the answer and its question may influence the answer’s perceived helpfulness.
Besides emotional content words, the language style (i.e., the usage of function words) of the text
is intricately linked to readers’ ability to access it and to evaluate its content. Numerous studies have
indicated that linguistic style matching (LSM), the congruence in the usage of function words, can
facilitate cognitive processing and improve fluency. For example, Zhang et al. (2021) demonstrated
that executives employing similar language styles (high LSM) in communications within their top
management teams were associated with more efficient corporate decision-making. Furthermore,
high LSM has been found to encourage individuals within a community to cultivate a common social
identity and minimize social distance in communication (Rains, 2016; Wang et al., 2019). However,
there appears to be a dearth of research on the impact of LSM on the perceived helpfulness of answers
within the context of online Q&A communities. To address this gap, this study aims to investigate
how LSM between the question and the answer in a thread influences the perceived helpfulness of
the answer.
The research framework for our study is depicted in Fig. 1. Drawing upon the theories and literature
pertaining to information-value assessment, this study probes the effects of emotional intensity and
question–answer congruence (emotional-intensity congruence and linguistic style matching) on the
helpfulness of answers. Specifically, emotional intensity is assumed to have an inverted U-shaped
Figure 1. Research Framework
Volume 32 • Issue 1
6
relationship with answer helpfulness, and a higher level of question–answer congruence is hypothesized
to positively relate to answer helpfulness.
Emotional intensity is defined as the proportion of expressed emotions within the answer content
(Jin et al., 2016; Peng et al., 2020; Zhou et al., 2023). When readers assess the helpfulness of an
answer, they are engaging in a decision-making activity (Jin & Liu, 2010). Readers are required to
process the information presented in the answer and determine whether the content is capable of
assisting them in resolving their issues (Garbarino & Edell, 1997; Peng et al., 2020). The literature
in decision science presents substantial evidence that affective cues (namely, emotional content)
significantly influence the final decision-making and outcomes (Toma & D’Angelo, 2015; Zhou et
al., 2023). Similarly, within the context of online health Q&A communities, it is logical to posit that
emotions embedded in answers will play a pivotal role in the decision-making process regarding
their helpfulness (Jin et al., 2016).
Research has indicated that emotional content can influence the perceived value of information
through arousal (Toma & D’Angelo, 2015; Yin et al., 2017). Arousal refers to the extent to which
a reader is energized by an experience. This arousal occurs when readers consider both verbal and
nonverbal emotional cues to ascertain the emotional state of the communicators (Toma & D’Angelo,
2015; Yin et al., 2017). Within online health Q&A communities, individuals who experience increased
emotional arousal are more inclined to deduce that the replier was endeavoring to establish an emotional
connection with the audience (Toma & D’Angelo, 2015; Yin et al., 2017) or to provide emotional
support (Jin et al., 2016). This implied effort can increase the readers’ perceived helpfulness value
of answers (Edwards, 2008; Yin et al., 2017). However, research has also pointed to a diminishing of
the positive correlation between emotional arousal and perceived helpfulness with increasing levels
of arousal (Yin et al., 2017). At higher and higher arousal levels, the inferred effort level becomes
less and less believable to the reader (Yin et al., 2017), prompting them to consider alternative
interpretations. For example, high levels of emotional arousal may lead readers to infer that a replier
was utilizing a more subjective versus objective writing style, was just “gushing” or “venting”
about his or her experience (Jensen et al., 2013; Park et al., 2007; Liu et al., 2020), or was imposing
unpleasant feelings on the reader (Friestad & Wright, 1994; Xu & Wyer, 2010; Toma & D’Angelo,
2015). Readers might perceive excessive levels of arousal induced by an abundance of emotional
content as an indication of irrationality (Pham, 2007; Fedorikhin & Patrick, 2010). Consequently, at
exceedingly high levels of expressed emotional content, the association between emotional content
and perceived helpfulness of answers may diminish or even invert.
Furthermore, research has demonstrated that emotional content is frequently processed
automatically, potentially leading readers to assimilate the content swiftly and enhance fluency (Park
et al., 2007; Li & Huang, 2020). The facilitation of information processing can increase readers’
perceived helpfulness value of content such as online reviews and health-related advice (Yin et al.,
2017; Li & Huang, 2020). Beyond emotional content, readers employ cognitive cues to deliberately
process incoming information and arrive at a more profound decision (Schwarz, 2000; Li & Huang,
2020). Although affective cues facilitate rapid information processing, deeper engagement with the
content necessitates cognitive cues, particularly when abundant information is available (Schwarz,
2000; Wang et al., 2017; Li & Huang, 2020). When cognitive cues are activated, they generate for the
reader more solid supporting reasons to make a certain decision (Li & Huang, 2020). If the answer
is overloaded with emotional content, other cognitive information resources will be crowded out. In
other words, if a replier expresses more emotional content in a fixed-length answer, less cognitive
information (such as specific methods, related theories, and so on) will appear in the answer. Thus,
an excessive inclusion of emotional content might undermine the perceived helpfulness of answers
by crowding out other information.
Volume 32 • Issue 1
7
To summarize, when readers evaluate answers and the emotional content is within a moderate
range, its amplified presence correlates with a higher likelihood of readers perceiving the replier’s
implied effort and expediting information processing, which in turn enhances the perceived helpfulness
of the answers (Toma & D’Angelo, 2015; Li & Huang, 2020). However, an excess of emotional
content in answers can result in a scarcity of cognitive cues for in-depth processing and may cultivate
an irrational and detached perception of the replier (Fedorikhin & Patrick, 2010; Li & Huang, 2020;
Liu et al., 2020), which in turn diminishes the perceived helpfulness of the answer. Therefore, we
hypothesize the following:
H1: The emotional intensity of an answer has an inverted U-shaped relationship with the perceived
helpfulness of the answer.
A growing body of research indicates that congruence between an entity and its contextual cues can
influence the perceived value of the entity (Edwards, 2008). For example, research in organization
science has demonstrated that when environmental factors align with individuals’ objectives, there is
often an enhancement in value assessments, such as the perceived value of their decisions (Higgins,
2000; Avnet & Higgins, 2006). In the context of UGC communities, research has underscored the
significance of congruence (such as word usage) between an entity and its contextual cues for content
helpfulness (Yin et al., 2016; Zhou et al., 2020; Yang et al., 2021). For example, in online review
platforms, the congruence between the rating of a focal review and the average rating of other reviews
can influence perceptions of review helpfulness (Yin et al., 2016). Similarly, within online Q&A
communities, when readers assess answers, they may be swayed by contextual cues, particularly the
question content in the corresponding thread. Question content serves as a crucial contextual cue, as
readers typically first read the question and assess its relevance to their inquiry prior to evaluating
the answers (Schwarz, 2000).
In our study, we argue that congruence in emotional intensity between question and answer
contributes to the perceived helpfulness of the answer. The first potential reason is that the congruence
facilitates processing and enhances fluency (Winkielman et al., 2012). After encountering questions
with a high emotional intensity, readers’ emotional mindsets are likely to be activated, leading to
a retention of emotional words in their memory (Wakslak & Trope, 2009). In other words, when
primed by questions of high emotional intensity, readers are expected to comprehend emotionally
charged answers more fluently than nonemotional ones. This fluency is thought to help readers
process answer information quickly and effortlessly, thereby increasing the perceived helpfulness of
the answer (Winkielman et al., 2012).
Additionally, when there is high congruence of emotional intensity between the question and
answer, it may result in the reader’s repeated exposure to affective stimuli. According to mere exposure
theory, repeated exposure to a stimulus generally enhances an individual’s attitude toward it, thereby
potentially increasing its perceived value (Zajonc, 1968; Bornstein & Agostino, 1992; Zhou et al.,
2020). These stimuli encompass words, images, and videos. Notably, emotional words represent a
significant factor in facilitating this repeated exposure (Bornstein & Agostino, 1992). Drawing on
this theory, Zhou et al. (2020) found that in online review communities, higher levels of title–content
emotional congruence are correlated with increased review helpfulness. In the context of our study,
we contend that repeated exposure to emotional words within questions and answers may enhance
readers’ preferences and, consequently, the perceived helpfulness of the answers.
Moreover, emotional words reflect the emotional processes that an individual undergoes (AL-
Smadi et al., 2017). In questions, more emotional words, particularly those with negative emotional
words, imply that the answer seeker may be experiencing serious and acute mental distress (Krueger
et al., 2001). When facing serious and acute conditions, individuals are inclined to conceptualize their
Volume 32 • Issue 1
8
ailments in tangible terms and desire more emotional support, as their capacity for rational thought
may be diminished. Emotional words are more vivid and concrete than nonemotional words, and
they can also provide emotional support. Readers of the question may be in the same situation as the
seeker, since it is often those who face similar dilemmas that seek out related questions (Schwarz,
2000). Consequently, when individuals encounter answers with emotional intensity corresponding
to the question’s level of seriousness and urgency, they are likely to experience enhanced cognitive
fluency, which can in turn elevate the perceived value of the help offered.
Above all, we hypothesize the following:
H2: The congruence of emotional intensity between the question and the answer in a thread is positively
related to the perceived helpfulness of the answer.
In addition to content words (such as emotional words), the other category of linguistic cues is function
words such as articles, pronouns, prepositions, and so on (Toma & D’Angelo, 2015). In text-based
communication, LSM is usually indicated by the congruence of function-word usage. Similar to the
congruence of emotional intensity, higher LSM is believed to facilitate information processing and
enhance fluency (Yin et al., 2017; Zhang et al., 2021). For example, high LSM between the manager’s
response and the review can help readers effectively understand the information provided by the
reviewer and then increase the perceived helpfulness of the review (Wang et al., 2019). Similarly, we
contend that high LSM between the question and the answer may also promote information processing
for readers, subsequently enhancing the perceived helpfulness of the answer.
In addition, high LSM indicates that language styles in conversation dyads are synchronized to a
greater degree, which can stimulate individuals in a community to form a common social identity and
reduce social distance in communication (Chung & Pennebaker, 2007; Anderson et al., 2008). Such
a perceived social identity can influence readers’ judgments and behaviors by providing heuristically
relevant diagnostic information (Chung & Pennebaker, 2007). For example, in online review platforms,
reviews with high LSM can help readers establish a social identity with the reviewer, which will
then increase the perceived helpfulness of the review and result in positive changes in the conversion
rate (Chung & Pennebaker, 2007; Wang et al., 2019). Aligned with our study, employing high LSM
promotes the development of a social identity among the readers of the answers. This social identity
might build the readers’ trust and subsequently elevate their perception of the answer helpfulness
(Schwarz, 2000; Jin & Liu, 2010). We hypothesize:
H3: The LSM between the question and the answer in a thread is positively related to the perceived
helpfulness of the answer.
Our research context is a leading online mental-health community in China, YiXinli (xinli001.com).
The reasons for selecting YiXinli as the context of this study are as follows. First, founded in 2011 by a
consortium of mental-health service providers, YiXinli was established with the aim of enhancing the
practice of mental-health services in China via the internet. As a forerunner in the domain of online
mental-health services, YiXinli has played a pivotal role in shaping the landscape of the industry.
By May 2020, the number of registered users had surpassed 22 million, thereby cementing its status
as one of the most expansive mental-health communities in China. The community’s sophisticated
Volume 32 • Issue 1
9
operational model and the high level of user engagement have rendered YiXinli an increasingly
prominent focus of academic inquiry within the field of online health communities (Zhou et al., 2019).
Second, while the public pays significant attention to physical-health issues, mental-health
services have remained a low priority in China for various reasons. In recent years, the problems
stemming from mental-health issues have intensified. Utilizing this mental-health Q&A community
as a research setting appears to have greater importance in addressing social problems that are
overlooked yet progressively damaging (Xiang et al., 2012).
Third, similar to other online health Q&A communities such as WebMD (Peng et al., 2020), the
community in our study also allows readers to locate and evaluate the answer based on a helpfulness
voting system. This online health Q&A community records each detail of the questions, answers,
and members (i.e., the seekers and repliers), such as concrete content, time stamp, and member level.
It lays the groundwork for acquiring valid data for our investigation into the factors influencing the
helpfulness of the answers. A screenshot of a Q&A session in the community is provided in Fig. 2,
with some necessary translation.
In January 2020, a total of 16,808 questions were collected, accompanied by 128,087
corresponding answers. In order to ensure the dataset’s quality and ensure the generalizability of
the study’s findings, we excluded answers from authors who participated only once, resulting in the
Figure 2. The Screenshot of a User Q&A Session
Volume 32 • Issue 1
10
removal of 7,011 samples. Therefore, the dataset utilized in this study comprises a total of 121,076
answers, contributed by 8,946 unique repliers. Among these answers, over 47% of them (56,906)
had received at least one helpfulness vote. The online health Q&A community provides us with
comprehensive information regarding the questions, answers, and individuals involved in questioning
and answering.
To be specific, the dataset comprises primarily two types of information: question information
and answer information, as shown in Fig. 2. Regarding question information, we gathered:
1) The question body. Furthermore, additional features such as readability and length can be derived
from the question body.
2) Classification of question types. The community organized questions into 12 distinct disease
topics for users to explore. These topics include interpersonal, marriage, family, love, emotion,
growth, treatment, hot spots, knowledge, professional, behavior, and dreams.
3) The reward amount allocated by the seeker. We can account for the impact of financial incentives,
as the external factor, on our empirical results.
4) Seeker information (such as whether the seeker is anonymous). The reason why we account for
that is similar to the reason why we consider whether the question is rewarded.
5) Publication time. We also gathered the specific time of the answer and can compute the time
duration between the question and the answer within a thread. In this manner, we can mitigate
the doubt surrounding the potential influence of time on our empirical results.
For answer information, we collected data on:
1) The number of helpfulness votes, which serves as a crucial measurement indicator for assessing
the perceived helpfulness of answers.
2) The answer content. We can rely on it (and text-processing tools can be utilized to extract
variables, including emotional intensity) for conducting empirical tests and providing support
for our theoretical model.
3) Answer time, which is illustrated as above.
4) Replier information (such as the replier’s reputation point provided by the community). It is helpful
to reduce the influence of individual-level factors (i.e., replier-level factors) in the empirical test.
Dependent Variable
The dependent variable in our study is answer helpfulness ( AnsHelpij ). Following Wang et al. (2019)
and Yang et al. (2021), we define AnsHelpij to denote the total number of helpfulness votes obtained
by answer
i
to question j.
Independent Variables
The independent variables in our study are emotional intensity, emotional-intensity congruence,
and LSM. To construct measures of these independent variables, we leveraged the latest version of
Linguistic Inquiry and Word Count (the Chinese version), a language-analysis tool for investigating
the relationship between word usage and psychological variables (Pennebaker et al., 2001). LIWC
has been widely used in psychology literature and recently has been increasingly used in information
systems and marketing literature (Yin et al., 2016; Wang et al., 2019), especially to quantify emotional
expression and LSM in various types of user-generated content (Huang et al., 2017; Peng et al., 2020;
Zhang et al., 2021).
Volume 32 • Issue 1
11
Emotional Intensity. We use AnsEmotionij to denote the emotional intensity of answer i to question
j. Following Yin et al. (2016), Wang et al. (2019), and Peng et al. (2020), we measure the
emotional intensity of the answer by calculating the number of emotional words (according to
the LIWC dictionary) divided by the total number of words.
Emotional-Intensity Congruence. QA Congruenceij
- is defined to denote the congruence of
emotional intensity of answer i to question j. Following Yin et al. (2016) and Peng et al.
(2020), we measure the congruence of the emotional-intensity variable through the commonly
used difference-score approach. The difference-score approach is deemed appropriate for
examining congruence concepts when congruence is defined as a match between two
component variables. The following are the concrete steps we took to measure the variable:
first, in a thread, calculate the emotional intensity of the question and its answers. Second,
standardize both components (the emotional intensity of the question and answers). Third,
calculate the absolute difference between these two standardized component variables.
Finally, multiply the absolute difference by -1 to ease interpretation. This measuring method
ensures that the higher value indicates the greater emotional-intensity congruence between
the answer and the question.
LSM. Following the studies of Schwarz (2000), Jin & Liu (2010), Toma & D’Angelo (2015), and
Wang et al. (2019), we measure LSM by calculating the degree of similarity of usage intensities
of function words between the answer and its question in the same thread. The linguistic style
matching of answer i to question j is denoted as LSMi j . The function word categories are
presented in Table 1.
Based on the above constructed function-word categories, the score of LSM can be calculated
in the following steps. First, collect the function-word lists from LIWC. Then, compute the function-
word usage intensity for each text (which included the question and answers in a thread) and for each
of nine function-word categories. Third, calculate nine LSM scores between each of the answers
and the relative question in the same thread for each function-word category. Finally, aggregate
the LSM scores obtained from the nine categories and then calculate the arithmetic average of
the LSM scores. In this way, we obtain the LSM scores as the similarity level of language styles
between an answer and its related question in a thread and use them in the later analyses. The
above steps are shown in Fig. 3.
Table 1. Function Words and Examples
Category Word Examples
Quantifiers few, many, much
Conjunctions and, but, whereas
Negations not, no, never
Prepositions after, in, with
Impersonal pronouns it, that, anything
Personal pronouns you, I, their
Common adverbs very, often, hardly
Articles a, an, the
Auxiliary verbs am, will, shall
Volume 32 • Issue 1
12
Control Variables
We also include a series of control variables to account for the influence of replier characteristics, answer
content characteristics, and question characteristics. Many studies have suggested that replier
characteristics, as the largest information source, can affect the perceived helpfulness of answers (Jin
et al., 2016; Zhang et al., 2020), so we account for replier characteristics. For the Q&A setting, (1) we
control Tenureik , indicating how long the replier
k
had been registered in the community when they
replied to the focal answer i. (2) We also account for PastContributionik , indicating the number of
replier
k
’s prior answers in the community at the time they replied to the focal answer
i
.
Answer content characteristics have also been found to contribute to information helpfulness
(Jin et al., 2016; Peng et al., 2020; Zhang et al., 2020). First, the literature has shown that answer
length can convey more information to readers (Jin et al., 2016). Thus, we control for answer length,
which is measured as the number of words in an answer, and define it as AnsLengthij , denoting the
length of answer i to question j. Second, the difficulty in reading an answer is a direct indication
of its perceived helpfulness (Jin et al., 2016; Zhang et al., 2020). Following other studies of online
Q&A communities that used words-to-sentences ratios to measure readability of answer text, we use
words-to-sentences ratios to control for the readability of the answer (the higher the ratios, the harder
it is to read) (Hofstetter et al., 2018; Yu et al., 2022). The readability of answer i to question j is
denoted as AnsReadabilityij . Third, to account for a potential duration effect, we also include the
time difference between the question post time and the answer post time measured in days (Lee et
al., 2019). Such time difference between answer
i
and question j is denoted as DifDaysij . Fourth,
to account for a potential competitive effect, we control for the total number of answers for a question,
and AnsSum j is defined to measure the answer sum of question j. Finally, to capture the differences
in the quantity or quality of the answers across question types (Hsieh et al., 2010; Peng et al., 2020),
we control for question types with a set of dummy variables based on the type of question. QueTagj
is defined to indicate which question type the question j belongs to.
Further, for question-related control variables, we control for an extensive list of such attributes.
First, we account for two variables that reflect the detailed degree of question content (Lee et al., 2019;
Zhang et al., 2020): SentenceSumj, representing the total number of sentences in the body of question
Figure 3. The Process of Calculating the LSM Score of Question and Answer
Volume 32 • Issue 1
13
j, and QueContentWordj, representing the total number of words in the content of question j
(normalized) (Peng et al., 2020). Second, to account for readers’ tendency to favor answers under paid
questions, we also consider whether a question had been offered a financial reward (Harper et al., 2008;
Jeon et al., 2010). We define Rewardj, which is a dummy variable, as equal to 1 if question j offers
a financial reward (and 0 otherwise). (3) Finally, considering the impact of question breadth (Harper et
al., 2008; Jeon et al., 2010), we control for the total number of subtags attached to a question, which is
defined as TagSumj, denoting the total number of subtags attached to question j.
The descriptive statistics for the key variables in our analysis are presented in Table 2. We
also display the correlation matrix of these variables in Table 3, along with the significance of the
correlation coefficients. Table 3 also provides the values for the variance inflation factor (VIF) values.
The coefficients do not appear to indicate a high correlation between the variables, whose threshold
is below 0.8, as suggested by Kline (2011). Furthermore, the correlation between the independent
variable and the dependent variables is significant, thereby providing preliminary validation for the
theoretical hypothesis mentioned above. Among all the VIF values, the maximum is 1.99, below
the conventional threshold of 10 (Liu et al., 2021), indicating no strong multicollinearity presents
in our study.
The dependent variable, AnsHelpij , is a nonnegative count variable with significant overdispersion
(as supported by a likelihood-ratio test for all regressions, p .<0 01 ) (Cameron & Trivedi, 1998;
Hendrickx, 2002). Given that traditional regression models, such as ordinary least squares, are biased
and inconsistent in this case, in line with Faraj et al. (2015) and Yang et al. (2019), we apply negative
binomial regression for the analysis.
To account for the replier heterogeneity, we employ replier-level fixed-effects negative binomial
regression. Meanwhile, we account for the heterogeneity in question types, which are algebraically
equivalent to including a dummy for each question type in our sample, as described previously (Peng
et al., 2020). Therefore, the overall model is shown in Equation 1.
AnsHelp Log AnsEmotion Log AnsEmotion LSM
ij ij ij
= +
( )
+
( )
+b b b b
2
2
3
)iij
+ − + + +β β α ε
4QA Congruence Controls
ij k ij (1)
wherein Log AnsEmotionij
( )
is natural log-transformed as Ln AnsEmotionij +
( )
1 (Chen et al.,
2021) and ( )Log AnsEmotionij
( )
2 is the square of Log AnsEmotionij
( )
.
Controls
is a set of control
variables as described previously, ak absorbs the replier-level fixed effect (FE), and eij is an
idiosyncratic error term. b b
0 4
- and
b
are the coefficients to be estimated.
Within the research framework of Equation 1, Table 4 presents the results of the hierarchical
approach to test our hypotheses. Model 1 is an estimation using only all control variables, and
emotional intensity is added to the existing variables in Model 2. Furthermore, Model 3 includes the
square of Log AnsEmotionij
( )
, that is, (Log AnsEmotionij
( )
2, to examine the nonlinear relationship
between the emotional intensity of the answer and answer helpfulness. Model 4 includes all three
focused variables to test our hypotheses.
Volume 32 • Issue 1
14
On the basis of Model 2, we conduct the likelihood ratio test to compare Model 3 with Model
2 and we find that the addition of the square of the Log AnsEmotionij
( )
term significantly
improves the model fit ( p<0 01. ). In Model 3 of Table 4, the coefficient of first term is
significantly positive ( bLog AnsEmotionij
( )
=3 701. , p<0 01. ), while the coefficient of second term
is significantly negative ( b
Log AnsEmotionij
( )
( )
= −
217 948. , p
<
0 01. ). This result supports our
Table 2. Variables and Descriptive Statistics
Variables Mean S.D. Min. Max.
Dependent Variable AnsHelpij 2.72 3.24 0 35
Independent Variables AnsEmotionij 0.07 0.04 0 1
LSMij 0.75 0.16 0.13 1.13
QA Congruenceij
--0.07 0.06 -1 0
Control Variables:
Answer characteristics AnsLengthij 332.04 291.71 38 5,447
AnsReadabilityij 0.02 0.02 0 0.48
AnsReadij 4,056.29 47,929.48 6 4,027,541
DifDaysij 8.10 38.33 0 691
AnsSumj7.62 10.37 1 133
Question characteristics Rewardj0.37 0.48 0 1
QueContentWordj0.38 0.25 0 1
QueSentSumj6.46 5.21 2 58
TagSumj2.22 0.79 0 3
QueTagj0 12
Replier characteristics PastContributionik 499.51 1,071.36 0 5,342
Tenureik 293.14 375.65 0 1,618
Volume 32 • Issue 1
15
Table 3. Correlation Coefficients and VIF Values (N = 121,076)
Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) VIF
AnsHelp
1.00
Log AnsEmotionij
( )
-0.08***
1.00
1.04
Lsmij
0.19***
0.004
1.00
1.28
QA Congruenceij
-
0.06***
-0.03***
0.20***
1.00
1.08
Log AnsLengthij
( )
0.60***
-0.15***
0.32***
0.07***
1.00
1.29
Log AnsReadabilityij
( )
-0.07***
-0.002
-0.10***
-0.03***
-0.10***
1.00
1.02
Log AnsSumj
( )
0.41***
0.02***
0.12***
0.08***
0.18***
-0.03***
1.00
1.81
Log AnsReadij
( )
0.31***
0.004
0.10***
0.09***
0.16***
-0.02***
0.61***
1.00
1.66
Log DifDaysij
( )
0.00***
-0.01**
-0.03***
0.01
0.04***
0.03***
0.10***
0.19***
1.00
1.04
Rewardj
0.27***
0.01***
0.17***
0.05***
0.20***
-0.05***
0.37***
0.26***
-0.004
1.00
1.26
QueContentWordj
0.16***
0.03***
0.34***
0.22***
0.18***
-0.03***
0.20***
0.15***
-0.002
0.26***
1.00
1.94
QueSentSumj
0.15***
-0.01
0.19***
0.14***
0.11***
-0.02***
0.21***
0.13***
0.01***
0.12***
0.61***
1.00
1.64
TagSumj
0.05***
0.03***
0.10***
-0.02***
0.09***
-0.02***
0.05***
0.01**
-0.02***
0.10***
0.17***
0.06***
1.00
1.05
Log PastContributionik
( )
-0.08***
0.005*
0.01***
0.01*
-0.10***
-0.04***
-0.11***
-0.03***
-0.10***
-0.03***
-0.01***
-0.01***
-0.05***
1.00
1.99
Log Tenureik
( )
0.06***
-0.04***
0.04***
0.01***
0.11***
-0.03***
-0.10***
-0.06***
-0.12***
-0.01***
0.00
0.00
-0.02***
0.67***
1.00
1.94
QueTagj
0.02***
-0.07***
-0.11***
-0.03***
-0.04***
0.02***
0.01***
-0.02***
0.04***
-0.06***
-0.16***
-0.03***
-0.06***
-0.02***
-0.01***
1.00
1.05
*p < 0.1; **p < 0.05; ***p < 0.01
Volume 32 • Issue 1
16
Table 4. The Results of Fixed-Effects Negative Binomial Model to Explore the Impact on the Number of Helpfulness Votes (N =
121,076)
AnsHelpij
Model 1 Model 2 Model 3 Model 4
Log AnsLengthij
( )
0.707*** (0.039) 0.710*** (0.042) 0.702*** (0.038) 0.701*** (0.039)
Log AnsReadabilityij
( )
-0.039** (0.014) -0.036** (0.016) -0.027* (0.014) -0.021 (0.014)
Log AnsSumj
( )
0.284*** (0.010) 0.283*** (0.012) 0.283*** (0.010) 0.282*** (0.012)
Log AnsReadij
( )
0.058*** (0.005) 0.059*** (0.005) 0.059*** (0.005) 0.059*** (0.005)
Log DifDaysij
( )
-0.062*** (0.012) -0.062*** (0.014) -0.062*** (0.010) -0.061*** (0.012)
Rewardj0.127*** (0.020) 0.127*** (0.020) 0.127*** (0.021) 0.126*** (0.017)
QueContentWordj0.082*** (0.012) 0.077*** (0.013) 0.073*** (0.012) 0.012 (0.015)
QueSentSumj0.003*** (0.0006) 0.003*** (0.0005) 0.003*** (0.0005) 0.003*** (0.0005)
TagSumj0.009** (0.004) 0.008** (0.004) 0.008* (0.004) 0.007* (0.004)
Log PastContributionik
( )
-0.010 (0.029) -0.010 (0.029) -0.010 (0.032) -0.010 (0.037)
Log Tenureik
( )
-0.003 (0.013) -0.003 (0.012) -0.003 (0.010) -0.002 (0.010)
Log AnsEmotionij
()0.829*** (0.161) 3.701*** (0.552) 3.173*** (0.490)
( )Log AnsEmotionij
( )
2-17.948*** (3.751) -15.134***
(3.409)
QA Congruenceij
-0.157*** (0.038)
LSMij 0.249*** (0.031)
Constant
-3.004***(0.283) -3.091***(0.313) -3.166***(0.254) -3.136*** (0.290)
continued on following page
Volume 32 • Issue 1
17
hypothesis H1 that the emotional intensity of an answer has an inverted U-shaped relationship
with the perceived helpfulness of the answer.
Next, as shown in Model 4 of Table 4, the congruence of emotional intensity is significantly
positively related to the helpfulness of the answer ( bLog QA Congruenceij
p
−
( )
= <0 157 0 01. , . ), suggesting
that the congruence of emotional intensity contributes to the perceived helpfulness of the answer.
This result supports our hypothesis H2, which suggests the emotional-intensity congruence between
the answer and the question in a thread can increase the perceived helpfulness of the answer.
The helpfulness of the answer shows a significantly positive coefficient on LSM (
bLSMij
p= <0 249 0 01. , . ), suggesting that a high LSM between the question and answer can increase
the perceived helpfulness of the answer. This result supports our hypothesis H3.
Regarding the nonlinear relationship between the emotional intensity of an answer and its
perceived helpfulness, using the significant coefficient of a quadratic term is a common method to
determine a nonlinear relationship. From an empirical standpoint, regression models may cause
overfitting, and this criterion is too weak. Following the three-step procedure proposed by Lind and
Mehlum (2010), we provide more rigorous evidence for the existence of an inverted U-shape to avoid
the aforementioned problem that may arise in the estimation (Haans et al., 2016). The results are
shown in Table 5. Table 5 indicates that the coefficient of the second term ( Log AnsEmotionij
( )
)
is significantly negative (Step 1), the slopes exhibit considerable opposite steepness at both ends of
the data range (Step 2), and both the turning point (0.105) and the 95% confidence interval of the
turning point fall comfortably within the data range (Step 3). Thus, the inverted U-shaped relationship
has been confirmed, indicating that the helpfulness of the answer initially increases with its emotional
intensity, but at a decreasing rate until it reaches a maximum. Subsequently, the helpfulness of the
answer decreases at an increasing rate. This result illustrates the confidence of our conclusion that
the emotional intensity of an answer has an inverted U-shaped relationship with the perceived
helpfulness of the answer.
To address potential endogeneity or overfitting issues in our study, we conduct a series of robustness
checks: (1) the combination with matching approach to avoid the potential bias caused by confounding
variables, (2) the alternative measure of replier characteristics to avoid possible errors caused by
variable selection, (3) the alternative regression model to test the solidity of our main test, and (4) the
Table 4. Continued
AnsHelpij
Model 1 Model 2 Model 3 Model 4
QueTagjControl Control Control Control
Replier FEs Yes Yes Yes Yes
-195,276.52 -195,230.93 -195,131.35 -195,023.25
Wald c25,699.73 6,927.05 10,374.72 7,833.94
Note. Bootstrap standard errors are in parentheses.
*p < 0.1; **p < 0.05; ***p < 0.01.
Volume 32 • Issue 1
18
zero-inflated negative binomial model to deal with possible bias due to the concentration of answer
helpfulness at zero values.
First, to mitigate potential bias arising from observable confounding variables in the selection
mechanism, we conduct a reexamination of our findings by integrating the matching method with
the main models previously outlined. To improve the reliability of our findings, we apply both
propensity score matching (PSM) and coarsened exact matching (CEM). This approach helps mitigate
the potential influence of model specification errors and ensures more robust results (Bapna et al.,
2019; Chen et al., 2020). The implementation details of the matching approaches are provided in the
Appendix. The estimation results are presented in Column 1 and Column 2 of Table 6, which are
consistent with the main analyses.
Second, in order to address possible errors arising from variable selection, we introduce another
control variable to account for the replier characteristics based on the original model. Answers
provided by officially endorsed repliers are generally perceived as more trustworthy by the audience,
thus contributing to a higher level of perceived helpfulness (Kumar et al., 2018). Thus, we
incorporate this characteristic of the replier by denoting Endorsementik , which is a dummy variable
equal to 1 if replier
k
had the official endorsement of the community when answering to the focal
answer
i
and equal to 0 otherwise. The result, as displayed in Column 3 of Table 6, aligns with
our primary examinations.
Third, in order to address the issue of interdependence among individual answers within the
same question type, which is a common problem of endogeneity, we employ a random coefficient
multilevel model to assess the robustness of our findings (Peng et al., 2020). In general, it has been
observed that answers that pertain to the same question type tend to exhibit greater similarity among
themselves compared to answers that do not pertain to the same question type. This phenomenon
gives rise to what is known as intraclass correlation (Kreft & de Leeuw, 1998; Klein & Kozlowski,
2000). In accordance with the findings of Peng et al. (2020), the utilization of the random coefficient
Table 5. Evidence of the Nonlinear Relationship Between Emotional Intensity and Answer Helpfulness
AnsHelpij
Step 1 Log AnsEmotionij
( )
3.173*** (0.490)
Log AnsEmotionij
( )
( )
2
-15.134*** (3.409)
Turning point 0.105
Step 2 Slope at Xl (min Log AnsEmotionij
( )
= 0) 3.160*** (6.890)
Slope at Xh (max Log AnsEmotionij
( )
= 0.6931) -17.608*** (-4.686)
Step 3 95% confidence interval, Fieller method [0.094, 0.129]
Sasabuchi test for U-shaped 4.69 ( p
<
0 01. )
X–Y relationship Inverted U-shaped
Note. T-statistics are in parentheses.
*p < 0.1; **p < 0.05; ***p < 0.01.
Volume 32 • Issue 1
19
Table 6. The Results of the Robustness Checks
AnswerHelpij
(1) (2) (3) (4) (5)
Log AnsLengthij
( )
0.712***
(0.033)
0.702***
(0.037)
0.698***
(0.040)
1.119***
(0.006)
0.962***
(0.004)
Log AnsReadabilityij
( )
-0.013
(0.017)
-0.031**
(0.016)
-0.021
(0.014)
-0.076***
(0.008)
-0.071***
(0.005)
Log AnsSumj
( )
0.281***
(0.012)
0.031***
(0.013)
0.281***
(0.012)
0.336***
(0.007)
0.334***
(0.004)
Log AnsReadij
( )
0.058***
(0.006)
0.061***
(0.007)
0.059***
(0.006)
0.105***
(0.004)
0.087***
(0.003)
Log DifDaysij
( )
-0.063***
(0.012)
-0.051**
(0.017)
-0.062***
(0.012)
-0.106***
(0.003)
-0.084***
(0.002)
Rewardj
0.129***
(0.017)
0.145***
(0.019)
0.125***
(0.017)
0.254***
(0.009)
0.214***
(0.006)
QueContentWordj
0.011
(0.015)
0.007
(0.014)
0.012
(0.015)
0.194***
(0.021)
-0.115***
(0.014)
QueSentSumj
0.003***
(0.001)
0.003***
(0.001)
0.003***
(0.0005)
0.0005
(0.001)
0.006***
(0.001)
TagSumj
0.006
(0.005)
0.008*
(0.004)
0.007*
(0.004)
-0.020***
(0.005)
-0.011**
(0.003)
Log PastContributionik
( )
-0.008
(0.030)
0.005
(0.040)
0.005
(0.044)
-0.010***
(0.002)
-0.006***
(0.001)
Log Tenureik
( )
-0.003
(0.010)
-0.007
(0.009)
0.047***
(0.003)
0.047***
(0.002)
Endorsementik
-0.417*
(0.252)
Log AnsEmotionij
(3.272***
(0.441)
3.118***
(0.494)
3.165***
(0.488)
2.799***
(0.396)
1.537***
(0.214)
Log AnsEmotionij
(
( )
2-16.046***
(2.964)
-15.136***
(3.078)
-14.997***
(3.389)
-10.983***
(2.342)
-8.943***
(1.208)
QA Congruenceij
-0.120**
(0.043)
0.186***
(0.053)
0.155***
(0.038)
0.540***
(0.072)
0.282***
(0.050)
LSMij
0.274***
(0.030)
0.239***
(0.027)
0.248***
(0.032)
0.577***
(0.030)
0.354***
(0.020)
continued on following page
Volume 32 • Issue 1
20
multilevel model allows for the consideration of the interdependence among individual answers within
the same question category. The estimation result of this model is presented in Column 4 of Table
6. By employing this alternative approach, we observe that the results align with the main analysis,
thus demonstrating the robustness of our findings.
Fourth, a portion of the answers in our dataset, specifically 27.9%, received zero helpfulness
votes. In order to better understand our theoretical development, we utilize the zero-inflated method
in addition to our main model. The zero-inflated method, as suggested by Miranda-Moreno and Fu
(2006) and Wang et al. (2015), is particularly effective in differentiating the causes of excess zeros.
Similarly, in order to account for the presence of concentration of zero values, we employ a replier-
level fixed-effects zero-inflated negative binomial regression model as suggested by Kumar et al.
(2018). The result is presented in Column 5 of Table 6, demonstrating conclusions that align with
those obtained from our primary examination.
Both emotional-intensity congruence and LSM refer to the aspect of congruence in language usage.
The former refers to emotional words (considered content words, indicative of social and psychological
concerns) (Toma & D’Angelo, 2015), and the latter refers to function words (regarded as relatively
content-free parts of sentences) (Toma & D’Angelo, 2015; Zhang et al., 2021). Exploring the joint
effect of both on the answer helpfulness will further enrich the literature on factors influencing the
answer helpfulness from the perspective of question–answer congruence.
As illustrated previously, congruence in language usage can facilitate information processing
and enhance cognitive fluency (Wakslak & Trope, 2009; Winkielman et al., 2012). A combination
congruence of different categories of words (emotional words and function words) could strengthen
Table 6. Continued
AnswerHelpij
(1) (2) (3) (4) (5)
LSMij ´
QA Congruenceij
-
1.108**
(0.387)
0.553**
(0.242)
0.573**
(0.234)
1.614***
(0.362)
0.620**
(0.267)
Constant
-3.228***
(0.251)
-2.979***
(0.264)
-3.111***
(0.287)
-8.541***
(0.054)
-5.870***
(0.038)
QueTagjControl Control Control Control Control
Replier FEs Yes Yes Yes Yes Yes
Observations 114,356 98,938 121,076 121,076 121,076
Log Likelihood -183,211.91 -155,143.611 -194,989.88 -244,867.21 -220,207.6
Wald c2
5,163.14 9,709.84 10,210.52 69,109.47 80,452.84
Note. Robust standard errors are reported in parentheses.
*p < 0.1, **p < 0.05, ***p < 0.01
Volume 32 • Issue 1
21
this effect to a greater degree than language congruence of a single category (i.e., either emotional
words or function words) (Wang et al., 2019). Thus, we predict a positive effect on the perceived
helpfulness when positive changes in emotion-intensity congruence combine with positive changes
in LSM. In other words, there is an interaction between emotional-intensity congruence and LSM,
such that positive changes in positive emotional-intensity congruence coupled with positive changes
in LSM lead to greater perceived helpfulness.
To test the above deduction and to aid interpretation of the joint effect, we add an interaction
term ( LSM QA Congruence
ij ij
× − ) to Equation 1. Before that, we center the two variables so that
we can directly interpret the coefficient of LSMij as the average effect evaluated at the means of
QA Congruenceij
- (Brown et al., 2007; Chen et al., 2020). The estimation model is shown in
Equation 2.
AnsHelp Log AnsEmotion Log AnsEmotion
ij ij ij
= +
( )
+
( )
+b b b b
0 1 2
2
3
( ) LLSMij
+ − + × − + + +β β β α ε
4 5
QA Congruence LSM QA Congruence Controls
ij ij ij k iij (2)
wherein dependent variable, independent variables, and control variables are the same as those in
Equation 1. We also control for the replier-level fixed effect, which is absorbed by ak, and eij is an
idiosyncratic error term.
The result is reported in Column 1 of Table 7, showing that the coefficients of the main
independent variables are consistent with those in Model 4 of Table 4 in terms of sign and significance.
Moreover, the coefficient of the interaction term LSM QA Congruence
ij ij
× − is significantly positive
(bLSM QA Congruence
ij ij
p
× − = <1 126 0 01. , . ), suggesting that the interaction between the increasing degrees
of LSM and the positive changes in congruence of emotional intensity significantly increases answer
helpfulness.
In our hypothesis development, we propose that a potential mechanism by which emotional intensity
influences the helpfulness of answers is through the relative amount of affective cues (i.e., emotional
content) and cognitive cues (i.e., cognitive content). To test this theoretical mechanism and enhance
the robustness of the conclusion that emotional intensity affects the helpfulness of the answers, we
construct another variable, denoted as Ratioij . Ratioij refers to the relative amount of affective cues
of answer
i
to question j, as shown in Equation 3.
Ratio Emotional Intensity
Emotional Intensity Cogni
ij
ij
ij
=+
ttive Intensityij
(3)
Then we examine the effect of such a constructed new variable on the helpfulness of answers;
the estimation framework is shown in Equation 4.
AnsHelp Ratio Ratio Controls
ij ij ij k ij
= + + + + +β β β β α ε
0 1 2
2, (4)
Volume 32 • Issue 1
22
Table 7. The Results of Fixed-Effects Negative Binomial Model to Explore the Impact on the Number of Helpfulness Votes (N =
121,076)
AnsHelpij
(1) (2)
Log AnsLengthij
( )
0.700*** (0.039) 0.684*** (0.042)
Log AnsReadabilityij
( )
-0.021 (0.017) -0.030** (0.013)
Log AnsSumj
( )
0.282*** (0.013) 0.284*** (0.011)
Log AnsReadij
( )
0.059*** (0.005) 0.058*** (0.004)
Log DifDaysij
( )
-0.061*** (0.013) -0.062*** (0.012)
Rewardj0.128*** (0.019) 0.127*** (0.017)
QueContentWordj0.013 (0.013) 0.080*** (0.013)
QueSentSumj0.003*** (0.0006) 0.003*** (0.0005)
TagSumj0.007 (0.004) 0.009** (0.004)
Log PastContributionik
( )
-0.010 (0.031) -0.010 (0.030)
Log Tenureik
( )
-0.002 (0.013) -0.003 (0.013)
Log AnsEmotionij
()3.153*** (0.481)
( )Log AnsEmotionij
( )
2-14.983*** (3.213)
Ratioij 0.777*** (0.106)
Ratioij
2-1.749*** (0.223)
QA Congruenceij
-0.137*** (0.047)
continued on following page
Volume 32 • Issue 1
23
wherein Ratioij
2 is the square of Ratioij .
Controls
are the same control variables as those in the
main test, ak is the replier-level fixed effect, and eij is an idiosyncratic error term.
The result is reported in Column 2 of Table 7, showing the inverted U-shape between the ratio
(i.e., the relative amount of affective cues) and the answer helpfulness (
b b
Ratio Ratio
p p= < = − <0 777 0 01 1 749 0 01
2
. , . ; . , . ). We also test the existence of the nonlinear
relationship and confirm this conclusion. The result is consistent with our hypothesized theoretical
model and supports our theoretical model of H1 again. This finding also suggests that the relative
amount of affective cues is one of the important potential mechanisms by which emotional intensity
affects answer usefulness.
Our study identifies four significant findings. First, drawing on 121,076 answers from a popular
online health Q&A community, empirical analyses reveal that emotional content correlates with
answer helpfulness in an inverted U-shaped relationship rather than in a linear relationship (i.e., in
a strictly positive or negative) or nonexistent relationship (Jin et al., 2016; Zhang et al., 2019; Li &
Huang, 2020). Further research has discovered that the proportional relationship between emotional
and cognitive content is the underlying mechanism leading to this inverted U-shaped relationship.
This finding further explains the literature on how emotional cues and cognitive cues interact to
influence judgments of information value (Schwarz, 2000; Wang et al., 2017; Li & Huang, 2020).
Second, we find that question content cues and answer content cues do not independently
contribute to answer helpfulness. Our findings demonstrate that question–answer congruence can have
Table 7. Continued
AnsHelpij
(1) (2)
LSMij 0.284*** (0.037)
LSM QA Congruence
ij ij
× − 1.126*** (0.424)
Constant -3.347*** (0.324) -3.135*** (0.266)
QueTagjControl Control
Replier FEs Yes Yes
Log Likelihood -195,018.31 -195,587.69
Wald c25,804.00 6,422.55
Note. Bootstrap standard errors are in parentheses.
*p < 0.1; **p < 0.05; ***p < 0.01.
Volume 32 • Issue 1
24
an impact on answer helpfulness. Specifically, we put forth and examine the impact of two distinct
types of question–answer congruence: emotion-intensity congruence and linguistic style matching. We
observe that emotion-intensity congruence can exert a positive influence on the perceived helpfulness
of answers. This phenomenon can be attributed to the fact that emotion-intensity congruence may
facilitate information processing and enhance fluency. This finding aligns with the exposure effect
theory, positing that repeated exposure to an object heightens an individual’s preference for it (Zajonc,
1968; Bornstein & Agostino, 1992; Zhou et al., 2020).
Third, we also discover that linguistic style matching between the question and answer is positively
correlated with answer helpfulness. The reason for the positive impact lies in the phenomenon of
verbal mimicry. When the question seeker and answer replier engage in verbal mimicry, they enhance
their social interaction, leading to more effective communication and information exchange. This
practice also fosters a shared social identity among answer repliers within a community, reducing the
social distance in their communication. These factors collectively influence readers’ perceptions of
answer helpfulness, ultimately leading to the observed positive correlation. This finding regarding
the efficacy of linguistic style matching in online health Q&A communities aligns with findings in
other online platforms such as online review platforms (Yin et al., 2017; Wang et al. 2019).
Last, these two forms of question–answer congruence, involving the congruence in different word
categories (i.e., emotional words and function words), can collectively influence answer helpfulness.
This could be attributed to the fact that a combination congruence of different word categories has
the potential to amplify the effect of facilitating information processing and enhancing fluency.
Consequently, this combined congruence exerts a more positive impact on answer helpfulness than
the congruence of a single word category (emotional words or function words) (Toma & D’Angelo,
2015; Wang et al., 2019).
This paper makes several theoretical contributions. First, our study contributes to the value judgment
of health information literature by deepening the understanding of how emotional content, considered
as one of the most significant content cues, affects the answer helpfulness value in online health
Q&A communities. The current literature has investigated the relationship between emotional content
and answer helpfulness value judgments. However, findings from existing empirical studies on the
effect of emotional content on perceived helpfulness present inconsistencies. Our study investigates
the role of emotional content in evaluating perceived helpfulness. The result shows that the effect of
emotional content on answer helpfulness, expected to follow an inverted U-shape, is more nuanced
than previously suggested by prior studies (Jin et al., 2016; Zhang et al., 2019; Li & Huang, 2020).
This nonmonotonic finding not only offers an explanation for the inconsistent findings regarding the
impact of emotional content on answer helpfulness in online health Q&A communities but also aligns
with conclusions derived from other user-generated content, such as online reviews (Yin et al., 2017).
Second, our research also augments the theoretical literature by investigating the role of question–
answer congruence in language attributes, thus broadening the study of factors that influence health-
related answer helpfulness. Prior research has commonly posited that the characteristics of answers
are the primary antecedents of perceived helpfulness and has taken into account context cues such
as question factors (Jin et al., 2016; Zhang et al., 2019). However, this research has focused solely
on the independent impacts of answer characteristics and question cues on answer helpfulness. Our
study advances a congruence perspective on question–answer language attribute congruence and
finds that two types of question–answer language-attribute congruence (emotion-intensity congruence
and linguistic style matching) can affect readers’ helpfulness evaluation, thereby unveiling a novel
perspective of studying influence factors of health-related answer helpfulness. Although our study
is based on an online health Q&A community, we believe that the congruence perspective can be
extended into other online Q&A communities and even more broadly into user-generated content
communities. Specifically, our findings on the impact of emotion-intensity congruence suggest
Volume 32 • Issue 1
25
that while emotional content may not invariably be beneficial, its influence can be modulated by
congruence with contextual cues, such as the emotional tone of the question. This finding offers an
alternative explanation for the inconsistent effects of emotional content.
Third, the conclusions of our research further enhance the literature on the role of linguistic style
matching in information value judgments. While some studies have recognized that high linguistic
style matching contributes to enhancing fluency and increasing the helpfulness value of text in
computer-mediated text-based communication (Yin et al., 2017; Zhang et al., 2021), there is a lack of
research on the effectiveness of linguistic style matching specifically within the context of online Q&A
communities. To our knowledge, our study is the first to examine the impact of linguistic style matching
and provide empirical evidence in the context of online health Q&A communities. Consequently, this
study contributes to the body of knowledge on linguistic style matching by affirming the significant
role of linguistic style matching in enhancing the helpfulness value of health-related answers.
Our study also provides several practical implications. First, our findings offer significant insights
for the design of online health Q&A communities. The provision of helpful content is crucial for the
growth and maintenance of online health Q&A community engagement, as users depend on such
content for decision-making. Consequently, platforms should implement guidelines to aid repliers in
crafting content that is more likely to be helpful (Lee et al., 2019; Zhang et al., 2020). Our research
outlines guidelines for the effective utilization of emotional content in answers. Platforms ought to
advise that repliers acknowledge that an insufficient or excessive amount of emotional content in an
answer may detract from increasing its perceived helpfulness and should therefore aim for a balanced
usage of emotional content. Additionally, rather than endorsing a simplistic formula for crafting helpful
answers, community managers are encouraged to guide repliers in tailoring their use of various word
categories in response to the language used in the question, including emotional and functional words.
Providing users with guidelines may be beneficial (Lee et al., 2019; Liu et al., 2020; Peng et al.,
2020; Zhang et al., 2020), but some repliers might not fully comprehend or effectively implement
the suggestions in these guidelines (Wang et al., 2022). A primary reason could be that repliers are
unable to calculate the language-feature intensity (such as emotional intensity) of answer texts in real
time and understand the optimal level of these features. To address these concerns, we suggest that
the platform, in addition to providing guidelines, should incorporate more design elements. First,
it might be beneficial if the managers could provide text examples. Text examples illustrating the
spectrum of emotional levels, ranging from “less than enough” to “appropriate” to “may be too many”
could be provided for repliers. Additionally, commercial products designed to assist with writing have
implemented linguistic-feature detectors (such as emotion), using dictionary methods and machine-
learning algorithms, to aid users in perceiving the tone of the text. Implementing similar automatic
linguistic-feature detectors in an online health community might be beneficial in helping repliers to
monitor the language-feature intensity of their answers in real time. As part of industrial validation,
we investigated whether UGC communities have any existing measures to enhance the effectiveness
of user-generated content. Research across various UGC community backgrounds (including online
health communities) has revealed an inverted U-shaped relationship between content length and
perceived helpfulness. Based on this conclusion, some online communities use the size of the
message box to suggest the optimal message length and encourage users to share suggestions. From
a design perspective, an online health community can provide suggestions to repliers, aiding them
in composing effective answers. Specifically, they could suggest an appropriate range of emotional
intensity and utilize sentiment-monitoring tools to guide repliers.
Second, our study provides critical insights into highlighting and locating more helpful answers
in online health Q&A communities. Identifying helpful answers can assist readers in reducing search
costs and enhancing decision-making efficiency. However, once an answer is posted, accumulating
helpfulness votes can be a time-intensive process. To facilitate quicker identification of helpful answers,
Volume 32 • Issue 1
26
many Q&A platforms have implemented predictive systems. Our study yields implications for the
selection of prediction features. In addition to question-and-answer content, the alignment between
the two—particularly in terms of emotional-intensity congruence and linguistic style matching—
warrants consideration in the estimation and prediction of answer helpfulness. Simultaneously, it is
imperative to consider the nonlinear effects of emotional content when identifying helpful answers.
Third, our study provides profound implications for the design of recommendation systems
within online health Q&A communities. Many health Q&A communities recommend questions to
repliers based on the degree of similarity between repliers’ expertise and the field of questions, a
method recognized as efficient in providing useful content. Our findings suggest that incorporating
considerations of linguistic style matching and emotional-language usage between repliers and seekers
into the design of recommendation systems could further incentivize repliers to contribute helpful
answers within online Q&A communities.
There are some limitations in our study, but they provide potential opportunities for future work in this
space. First, our study did not investigate all the underlying mechanisms that explain how emotional
content and question–answer congruence affect answer helpfulness. Future research can explore these
mechanisms using surveys or field experiments. Second, although we have selected as many control
variables as possible and have conducted a series of robustness tests, some endogeneity concerns still
remain. Future research could apply randomized field experiments to solve these endogeneity problems
and examine this causal influence. Third, although our study focuses on the impact of emotional
content and question–answer congruence in the context of online health Q&A communities, these
effects may also exist in other domains. Future research can explore the role of emotional content and
content-context congruence in other UGC communities. Fourth, we focus only on emotional-intensity
congruence and LSM in the present study. These two variables are theory-driven, but they have been
demonstrated as critical factors in readers’ value judgments. From the congruence perspective, other
aspects can also be considered in future research.
This work was supported by Program for Innovative Research Team of Shanghai University of Finance
and Economics (IRTSHUFE).
We declare that we have no conflict of interest.
Received: October 18, 2022, Revision: March 19, 2024, Accepted: March 29, 2024
Correspondence should be addressed to Xiaohang Zhou; xiaohang950510@163.com
Volume 32 • Issue 1
27
Adamic, L. A., Zhang, J., Bakshy, E., & Ackerman, M. S. (2008). Knowledge sharing and yahoo answers:
Everyone knows something. In WWW ’08: Proceeding of the 17th international conference on World Wide Web
(pp. 665–674). ACM. doi:10.1145/1367497.1367587
AL-Smadi, M., Jaradat, Z., AL-Ayyoub, M., & Jararweh, Y.AL-Smadi. (2017). Paraphrase identification and
semantic text similarity analysis in Arabic news tweets using lexical, syntactic, and semantic features. Information
Processing & Management, 53(3), 640–652. doi:10.1016/j.ipm.2017.01.002
Anderson, B., Goldin, P. R., Kurita, K., & Gross, J. J. (2008). Self-representation in social anxiety disorder:
Linguistic analysis of autobiographical narratives. Behaviour Research and Therapy, 46(10), 1119–1125.
doi:10.1016/j.brat.2008.07.001
Avnet, T., & Higgins, E. T. (2006). How regulatory fit affects value in consumer choices and opinions. JMR,
Journal of Marketing Research, 43(1), 1–10. doi:10.1509/jmkr.43.1.1
Bae, B. J., & Yi, Y. J. (2017). What answers do questioners want on social Q&A? User preferences of answers
about STDs. Internet Research, 27(5), 1104–1121. doi:10.1108/IntR-08-2016-0245
Bapna, S., Benner, M. J., & Qiu, L. (2019). Nurturing online communities: An empirical investigation.
Management Information Systems Quarterly, 43(2), 425–452. doi:10.25300/MISQ/2019/14530
Bornstein, R. F., & D’Agostino, P. R. (1992). Stimulus recognition and the mere exposure effect. Journal of
Personality and Social Psychology, 63(4), 545–552. doi:10.1037/0022-3514.63.4.545
Brown, J., Broderick, A. J., & Lee, N. (2007). Word of mouth communication within online communities:
Conceptualizing the online social network. Journal of Interactive Marketing, 21(3), 2–20. doi:10.1002/dir.20082
Cameron, A. C., & Trivedi, P. K. (1998). Regression analysis of count data. Cambridge University Press.,
doi:10.1017/CBO9780511814365
Chen, L., Xu, P., & Liu, D. (2020). Effect of crowd voting on participation in crowdsourcing contests. Journal
of Management Information Systems, 37(2), 510–535. doi:10.1080/07421222.2020.1759342
Chen, P. Y., Pavlou, P., Wu, S., & Yang, Y. (2021). Attracting high-quality contestants to contest in the context
of crowdsourcing contest platform. Production and Operations Management, 30(6), 1751–1771. doi:10.1111/
poms.13340
Chen, Y., Ho, T. H., & Kim, Y. M. (2010). Knowledge market design: A field experiment at Google Answers.
Journal of Public Economic Theory, 12(4), 641–664. doi:10.1111/j.1467-9779.2010.01468.x
Cheng, L. C., & Yang, Y. (2022). The effect of online reviews on movie box office sales: An integration of
aspect-based sentiment analysis and economic modeling. [JGIM]. Journal of Global Information Management,
30(1), 1–16. doi:10.4018/JGIM.298652
Cheung, C. M. Y., Sia, C. L., & Kuan, K. K. Y. (2012). Is this review believable? A study of factors affecting
the credibility of online consumer reviews from an ELM perspective. Journal of the Association for Information
Systems, 13(8), 618–635. doi:10.17705/1jais.00305
Chung, C. K., & Pennebaker, J. W. (2007). The psychological functions of function words. In K. Fiedler (Ed.),
Social communication (pp. 343–359). Psychology Press.
Chung, S., & Park, J. (2013). Effects of social and temporal distance on evaluation of corporate ambivalent
behavior. Social Behavior and Personality, 41(7), 1219–1224. doi:10.2224/sbp.2013.41.7.1219
Cronin, R. M., Conway, D., Condon, D., Jerome, R. N., Byrne, D. W., & Harris, P. A. (2018). Patient and
healthcare provider views on a patient-reported outcomes portal. Journal of the American Medical Informatics
Association : JAMIA, 25(11), 1470–1480. doi:10.1093/jamia/ocy111
Edwards, J. R. (2008). Person–environment fit in organizations: An assessment of theoretical progress. The
Academy of Management Annals, 2(1), 167–230. doi:10.5465/19416520802211503
Eslami, S. P., Ghasemaghaei, M., & Hassanein, K. (2018). Which online reviews do consumers find most
helpful? A multi-method investigation. Decision Support Systems, 113, 32–42. doi:10.1016/j.dss.2018.06.012
Volume 32 • Issue 1
28
Fang, C., & Zhang, J. (2019). Users’ continued participation behavior in social Q&A communities: A motivation
perspective. Computers in Human Behavior, 92, 87–109. doi:10.1016/j.chb.2018.10.036
Faraj, S., Kudaravalli, S., & Wasko, M. (2015). Leading collaboration in online communities. Management
Information Systems Quarterly, 39(2), 393–412. doi:10.25300/MISQ/2015/39.2.06
Fedorikhin, A., & Patrick, V. M. (2010). Positive mood and resistance to temptation: The interfering influence
of elevated arousal. The Journal of Consumer Research, 37(4), 698–711. doi:10.1086/655665
Filieri, R., Raguseo, E., & Vitari, C. (2018). When are extreme ratings more helpful? Empirical evidence on
the moderating effects of review characteristics and product type. Computers in Human Behavior, 88, 134–142.
doi:10.1016/j.chb.2018.05.042
Finney Rutten, L. J., Blake, K. D., Greenberg-Worisek, A. J., Allen, S. V., Moser, R. P., & Hesse, B. W. (2019).
Online health information seeking among US adults: Measuring progress toward a healthy people 2020 objective.
Public Health Reports (Washington, D.C.), 134(6), 617–625. doi:10.1177/0033354919874074
Friestad, M., & Wright, P. (1994). The persuasion knowledge model: How people cope with persuasion attempts.
The Journal of Consumer Research, 21(1), 1–31. doi:10.1086/209380
Garbarino, E. C., & Edell, J. A. (1997). Cognitive effort, affect, and choice. The Journal of Consumer Research,
24(2), 147–158. doi:10.1086/209500
Haans, R. F. J., Pieters, C., & He, Z. (2016). Thinking about U: Theorizing and testing U- and inverted U-shaped
relationships in strategy research. Strategic Management Journal, 37(7), 1177–1195. doi:10.1002/smj.2399
Ham, J., Lee, K., Kim, T., & Koo, C. (2019). Subjective perception patterns of online reviews: A comparison
of utilitarian and hedonic values. Information Processing & Management, 56(4), 1439–1456. doi:10.1016/j.
ipm.2019.03.011
Harper, F. M., Raban, D., Rafaeli, S., & Konstan, J. A. (2008). Predictors of answer quality in online Q&A sites.
In CHI ’08: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 865–874).
ACM. doi:10.1145/1357054.1357191
Harris Interactive. (2011, September 15). The growing influence and use of health care information obtained
online. Retrieved November 12, 2021 from https://www.prnewswire.com/news-releases/the-growing-influence-
and-use-of-health-care-information-obtained-online-129893198.html
Heinrich, C., Maffioli, A., & Vázquez, G. (2010). A primer for applying propensity-score matching. Inter-
American Development Bank, Technical Note No. IDB-TN-161. https://publications.iadb.org/en/primer-applying-
propensity-score-matching
Hendrickx, J. (2002). Review of regression models for categorical dependent variables using Stata by Long and
Freese. The Stata Journal, 2(1), 103–105. doi:10.1177/1536867X0200200107
Hesse, B. W., Nelson, D. E., Kreps, G. L., Croyle, R. T., Arora, N. K., Rimer, B. K., & Viswanath, K. (2005).
Trust and sources of health information: The impact of the internet and its implications for health care providers:
Findings from the first health information national trends survey. Archives of Internal Medicine, 165(22),
2618–2624. doi:10.1001/archinte.165.22.2618
Higgins, E. T. (2000). Making a good decision: Value from fit. The American Psychologist, 55(11), 1217–1230.
doi:10.1037/0003-066X.55.11.1217
Hofstetter, R., Zhang, J. Z., & Herrmann, A. (2018). Successive open innovation contests and incentives: Winner-
take-all or multiple prizes? Journal of Product Innovation Management, 35(4), 492–517. doi:10.1111/jpim.12424
Hsieh, G., Kraut, R. E., & Hudson, S. E. (2010). Why pay?: Exploring how financial incentives are used for
question & answer. In CHI ’10: Proceedings of the SIGCHI International Conference on Human Factors in
Computing Systems (pp. 305–314). ACM. doi:10.1145/1753326.1753373
Huang, N., Hong, Y., & Burtch, G. (2017). Social network integration and user content generation: Evidence
from natural experiments. Management Information Systems Quarterly, 41(4), 1035–1058. doi:10.25300/
MISQ/2017/41.4.02
Volume 32 • Issue 1
29
Huang, S., & Wang, L. (2022). An exploration study on the consequence of the COVID-19 pandemic on online
Q&A communities. Journal of Global Information Management, 30(1), 1–22. doi:10.4018/JGIM.301603
Iacus, S. M., King, G., & Porro, G. (2011). Multivariate matching methods that are monotonic imbalance
bounding. Journal of the American Statistical Association, 106(493), 345–361. doi:10.1198/jasa.2011.tm09599
Jensen, M. L., Averbeck, J. M., Zhang, Z., & Wright, K. B. (2013). Credibility of anonymous online product
reviews: A language expectancy perspective. Journal of Management Information Systems, 30(1), 293–324.
doi:10.2753/MIS0742-1222300109
Jeon, G. Y. J., Kim, Y. M., & Chen, Y. (2010). Re-examining price as a predictor of answer quality in an online
Q&A site. In CHI ’10: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp.
325–328). ACM. doi:10.1145/1753326.1753375
Jin, J., & Liu, Y. (2010). How to interpret the helpfulness of online product reviews: Bridging the needs between
customers and designers. In SMUC ’10: Proceedings of the 2nd International Workshop on Search and Mining
User-Generated Contents (pp. 87–94). ACM. doi:10.1145/1871985.1872000
Jin, J., Yan, X., Li, Y., & Li, Y. (2016). How users adopt healthcare information: An empirical study of an online
Q&A community. International Journal of Medical Informatics, 86, 91–103. doi:10.1016/j.ijmedinf.2015.11.002
Jones, Q., Ravid, G., & Rafaeli, S. (2004). Information overload and the message dynamics of online interaction
spaces: A theoretical model and empirical exploration. Information Systems Research, 15(2), 194–210.
doi:10.1287/isre.1040.0023
Kazmer, M. M., Lustria, M. L. A., Cortese, J., Burnett, G., Kim, J.-H., Ma, J., & Frost, J. (2014). Distributed
knowledge in an online patient support community: Authority and discovery. Journal of the Association for
Information Science and Technology, 65(7), 1319–1334. doi:10.1002/asi.23064
Kim, S., Oh, J. S., & Oh, S. (2007). Best-answer selection criteria in a social Q&A site from the user-oriented
relevance perspective. Proceedings of the American Society for Information Science and Technology, 44(1),
1–15. doi:10.1002/meet.1450440256
Kim, S., & Oh, S. (2009). Users’ relevance criteria for evaluating answers in a social Q&A site. Journal of the
American Society for Information Science and Technology, 60(4), 716–727. doi:10.1002/asi.21026
Klein, K. J., & Kozlowski, S. W. J. (Eds.). (2000). Multilevel theory, research, and methods in organizations:
Foundations, extensions, and new directions (pp. 3–90). Jossey-Bass/Wiley.
Kline, R. B. (2011). Convergence of structural equation modeling and multilevel modeling. In The SAGE handbook
of innovation in social research methods (pp. 562–589). SAGE Publications., doi:10.4135/9781446268261.n31
Kreft, I. G., & de Leeuw, J. (1998). Introducing multilevel modeling. SAGE Publications., doi:10.4135/9781849209366
Krueger, R. F., McGue, M., & Iacono, W. G. (2001). The higher-order structure of common DSM mental disorders:
Internalization, externalization, and their connections to personality. Personality and Individual Differences,
30(7), 1245–1259. doi:10.1016/S0191-8869(00)00106-9
Kumar, N., Qiu, L., & Kumar, S. (2018). Exit, voice, and response on digital platforms: An empirical investigation
of online management response strategies. Information Systems Research, 29(4), 849–870. doi:10.1287/
isre.2017.0749
Lee, S. Y., Rui, H., & Whinston, A. B. (2019). Is best answer really the best answer? The politeness bias.
Management Information Systems Quarterly, 43(2), 579–600. doi:10.25300/MISQ/2019/14160
Li, M., & Huang, P. (2020). Assessing the product review helpfulness: Affective-cognitive evaluation and
the moderating effect of feedback mechanism. Information & Management, 57(7), 103359. doi:10.1016/j.
im.2020.103359
Lind, J. T., & Mehlum, H. (2010). With or without U? The appropriate test for a U-shaped relationship. Oxford
Bulletin of Economics and Statistics, 72(1), 109–118. doi:10.1111/j.1468-0084.2009.00569.x
Liu, X., Wang, G. A., Fan, W., & Zhang, Z. (2020). Finding useful solutions in online knowledge communities:
A theory-driven design and multilevel analysis. Information Systems Research, 31(3), 731–752. doi:10.1287/
isre.2019.0911
Volume 32 • Issue 1
30
Liu, Z., Hatton, M. R., Kull, T., Dooley, K., & Oke, A. (2021). Is a large award truly attractive to solvers? The
impact of award size on crowd size in innovation contests. Journal of Operations Management, 67(4), 420–449.
doi:10.1002/joom.1132
Lou, J., Fang, Y., Lim, K. H., & Peng, J. Z. (2013). Contributing high quantity and quality knowledge to online
Q&A communities. Journal of the American Society for Information Science and Technology, 64(2), 356–371.
doi:10.1002/asi.22750
Miranda-Moreno, L. F., & Fu, L. (2006). A comparative study of alternative model structures and criteria for ranking
locations for safety improvements. Networks and Spatial Economics, 6(2), 97–110. doi:10.1007/s11067-006-7695-2
Mudambi, S. M., & Schuff, D. (2010). What makes a helpful online review? A study of customer reviews on
Amazon.com. Management Information Systems Quarterly, 34(1), 185–200. doi:10.2307/20721420
Oh, S., & Worrall, A. (2013). Health answer quality evaluation by librarians, nurses, and users in social Q&A.
Library & Information Science Research, 35(4), 288–298. doi:10.1016/j.lisr.2013.04.007
Park, D. H., Lee, J., & Han, I. (2007). The effect of on-line consumer reviews on consumer purchasing intention:
The moderating role of involvement. International Journal of Electronic Commerce, 11(4), 125–148. doi:10.2753/
JEC1086-4415110405
Peng, C. H., Yin, D., & Zhang, H. (2020). More than words in medical question-and-answer sites: A content-
context congruence perspective. Information Systems Research, 31(3), 913–928. doi:10.1287/isre.2020.0923
Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2001). Linguistic inquiry and word count. Psychology Press.
Pham, L., Klaus, T., & Changchit, C. (2023). Factors influencing intention to use online consumer reviews: The
case of Vietnam. [JGIM]. Journal of Global Information Management, 31(1), 1–22. doi:10.4018/JGIM.321642
Pham, M. T. (2007). Emotion and rationality: A critical review and interpretation of empirical evidence. Review
of General Psychology, 11(2), 155–178. doi:10.1037/1089-2680.11.2.155
Rains, S. A. (2016). Language style matching as a predictor of perceived social support in computer-
mediated interaction among individuals coping with illness. Communication Research, 43(5), 694–712.
doi:10.1177/0093650214565920
Schwarz, N. (2000). Emotion, cognition, and decision making. Cognition and Emotion, 14(4), 433–440.
doi:10.1080/026999300402745
Shah, C., & Pomerantz, J. (2010). Evaluating and predicting answer quality in community QA. In SIGIR ’10:
Proceedings of the 33rd International ACM SIGIR Conference on Research and Development in Information
Retrieval (pp. 411–418). ACM. doi:10.1145/1835449.1835518
Silberg, W. M., Lundberg, G. D., & Musacchio, R. A. (1997). Assessing, controlling, and assuring the quality
of medical information on the internet. Journal of the American Medical Association, 277(15), 1244–1245.
doi:10.1001/jama.1997.03540390074039
Tan, S. S., & Goonawardene, N. (2017). Internet health information seeking and the patient-physician relationship:
A systematic review. Journal of Medical Internet Research, 19(1), e9. doi:10.2196/jmir.5729
Toma, C. L., & D’Angelo, J. D. (2015). Tell-tale words: Linguistic cues used to infer the expertise of online
medical advice. Journal of Language and Social Psychology, 34(1), 25–45. doi:10.1177/0261927X14554484
Wakslak, C., & Trope, Y. (2009). The effect of construal level on subjective probability estimates. Psychological
Science, 20(1), 52–58. doi:10.1111/j.1467-9280.2008.02250.x
Wang, H., Ji, Y., & Wang, W. (2017). Quality evaluation for answers in Chinese QA community based on random forest––
Examples from Zhihu. In Proceedings of the 2017 International Conference on Humanities Science, Management and
Education Technology (HSMET 2017) (pp. 1184–1190). Atlantis Press. doi:10.2991/hsmet-17.2017.218
Wang, J., Gupta, M., & Rao, H. R. (2015). Insider threats in a financial institution: Analysis of attack-proneness of information
systems applications. Management Information Systems Quarterly, 39(1), 91–112. doi:10.25300/MISQ/2015/39.1.05
Wang, X., Jiang, M., Han, W., & Qiu, L. (2022). Do emotions sell? The impact of emotional expressions on sales
in the space-sharing economy. Production and Operations Management, 31(1), 65–82. doi:10.1111/poms.13527
Volume 32 • Issue 1
31
Wang, X., Tang, L. R., & Kim, E. (2019). More than words: Do emotional content and linguistic style matching
matter on restaurant review helpfulness? International Journal of Hospitality Management, 77, 438–447.
doi:10.1016/j.ijhm.2018.08.007
Winkielman, P., Huber, D. E., Kavanagh, L. C., & Schwarz, N. (2012). Fluency of consistency: When thoughts
fit nicely and flow smoothly. In B. Gawronski & F. Strack (Eds.), Cognitive consistency: A fundamental principle
in social cognition. Guilford Press.
Xiang, Y. T., Yu, X., Sartorius, N., Ungvari, G. S., & Chiu, H. F. K. (2012). Mental health in China: Challenges
and progress. Lancet, 380(9855), 1715–1716. doi:10.1016/S0140-6736(11)60893-3
Xu, A. J., & Wyer, R. S. Jr. (2010). Puffery in advertisements: The effects of media context, communication
norms, and consumer knowledge. The Journal of Consumer Research, 37(2), 329–343. doi:10.1086/651204
Yan, Z., Kuang, L., & Qiu, L. (2022). Prosocial behaviors and economic performance: Evidence from an
online mental healthcare platform. Production and Operations Management, 31(10), 3859–3876. doi:10.1111/
poms.13792
Yang, M., Ren, Y., & Adomavicius, G. (2019). Understanding user-generated content and customer engagement
on Facebook business pages. Information Systems Research, 30(3), 839–855. doi:10.1287/isre.2019.0834
Yang, S., Zhou, C., & Chen, Y. (2021). Do topic consistency and linguistic style similarity affect online review
helpfulness? An elaboration likelihood model perspective. Information Processing & Management, 58(3),
102521. doi:10.1016/j.ipm.2021.102521
Yin, D., Bond, S. D., & Zhang, H. (2017). Keep your cool or let it out: Nonlinear effects of expressed arousal on
perceptions of consumer reviews. JMR, Journal of Marketing Research, 54(3), 447–463. doi:10.1509/jmr.13.0379
Yin, D., Mitra, S., & Zhang, H. (2016). Research note—When do consumers value positive vs. negative reviews?
An empirical investigation of confirmation bias in online word of mouth. Information Systems Research, 27(1),
131–144. doi:10.1287/isre.2015.0617
Yu, Y., Khern-am-nuai, W., & Pinsonneault, A. (2022). When paying for reviews pays off: The case of
performance-contingent monetary rewards. Management Information Systems Quarterly, 46(1), 609–626.
doi:10.25300/MISQ/2022/15488
Zajonc, R. B. (1968). Attitudinal effects of mere exposure. Journal of Personality and Social Psychology, 9(2),
1–27. doi:10.1037/h0025848
Zhang, T., Liu, F. C., Gao, B., & Yen, D. (2021). Top management team social interaction and conservative
reporting decision: A language style matching approach. Decision Support Systems, 142, 113469. doi:10.1016/j.
dss.2020.113469
Zhang, Y., Li, X., & Fan, W. (2020). User adoption of physician’s replies in an online health community:
An empirical study. Journal of the Association for Information Science and Technology, 71(10), 1179–1191.
doi:10.1002/asi.24319
Zhang, Y., Lu, T., Phang, C. W., & Zhang, C. (2019). Scientific knowledge communication in online Q&A
communities: Linguistic devices as a tool to increase the popularity and perceived professionalism of knowledge
contribution. Journal of the Association for Information Systems, 20(8), 1129–1173. doi:10.17705/1jais.00563
Zhang, Y., & Wang, P. (2016). Interactions and user-perceived helpfulness in diet information social questions
& answers. Health Information and Libraries Journal, 33(4), 295–307. doi:10.1111/hir.12158
Zhou, J., Zhang, Q., Zhou, S., Li, X., & Zhang, X. (2023). Unintended emotional effects of online health
communities: A text mining-supported empirical study. Management Information Systems Quarterly, 47(1),
195–226. doi:10.25300/MISQ/2022/17018
Zhou, J., Zuo, M., & Ye, C. (2019). Understanding the factors influencing health professionals’ online voluntary
behaviors: Evidence from YiXinLi, a Chinese online health community for mental health. International Journal
of Medical Informatics, 130, 103939. doi:10.1016/j.ijmedinf.2019.07.018
Zhou, Y., Yang, S., Li, Y., Chen, Y., Yao, J., & Qazi, A. (2020). Does the review deserve more helpfulness when
its title resembles the content? Locating helpful reviews by text mining. Information Processing & Management,
57(2), 102179. doi:10.1016/j.ipm.2019.102179
Volume 32 • Issue 1
32
As mentioned in Literature Review section, Table 8 in the Appendix summarizes the examined
influence factors of the perceived value of answers in online Q&A communities.
Table 8. Literature About Factors of Perceived Value of Answers in Online Q&A Communities
Literature Answer Factors Question Factors Replier Factors
Kim et al. (2007)
Accuracy completeness;
characteristics;
cognitive value;
emotion support
Source quality
Admic et al. (2008) Answer length Repliers’ expertise
Harpar (2008) Question type;
reward
Kim and Oh (2009)
Accuracy;
length;
utility;
cognitive value;
emotion support
Chen et al. (2010) Reward Repliers’ reputation
Shah and Pomerantz (2010)
Enough information;
polite;
readability;
objective
…..
Expertise
Lou et al. (2013) Internal motivation;
external motivation
Oh and Worrall (2013)
Accuracy;
objectivity;
confidence;
effort
Source credibility
Toma and D’Angelo (2015)
Answer length;
concreteness;
psychological distancing
Zhang and Wang (2016)
Objectivity;
socio-emotional value
….
Expert
Jin et al. (2016)
Information quality;
emotion support
…..
Source credibility
….
Bae and Yi (2017)
Length;
numeric information;
references
Expert
Lee et al. (2019) Politeness Status
Zhang et al. (2019)
Readability;
completeness;
objective
….
Online experience;
offline expertise
Zhang et al. (2020)
Typographical cues;
metaphor;
humor;
confidence;
examples
Liu et al. (2020)
Appropriate amount;
ease of understanding;
objectivity;
timeliness
…..
Past experience;
expertise
……
Volume 32 • Issue 1
33
In the matching process, we divide the emotional intensity of answer
i
to question j into two
groups based on the median, and we label the group below the median as weak emotion and the other
group as strong emotion. Before the PSM, we randomize the order of samples to ensure that the
matching process is not affected by their order. We use the nearest-neighbor matching algorithm to
match in terms of the answer characteristics and question characteristics as shown in Table 9 (Heinrich
et al., 2010; Bapna et al., 2019). We perform t-tests of equality of means before (the left panel in
Table 9) and after (the right panel in Table 9) the matching so that we can check whether observable
characteristics are balanced across posts in the treatment and control groups. As shown in Table 9,
after PSM, most of the differences of means are no longer statistically significant, suggesting that
PSM helped reduce the bias associated with differences in observable characteristics. Fig. 4 shows
the standardized percentage bias for each covariate before and after PSM, which is consistent with
Table 9. Fig. 5 shows the propensity score histogram for the treatment and control groups (Heinrich
et al., 2010; Bapna et al., 2019). As shown in Fig. 5, the distribution between the two groups reveals
a clear region of common support. As mentioned above, the quality of the PSM matching method is
high enough that the test results of robust tests using the subsamples obtained by PSM are reliable.
The CEM procedure greatly reduces the global multivariate imbalance score, the L1 statistic
between the weak-emotion and strong-emotion groups, from 0.997 to 0.976, indicating a successful
matching (Iacus et al., 2011). Table 10 presents evidence that the CEM procedure has resulted in a
better balance between treatment and control groups in terms of mean differences of these two groups
and multivariate L1 distance.
Table 9. Differences in Mean Before and After PSM
Before PSM After PSM
Mean Weak
Emotion
Mean
Strong
Emotion
%Bias t p
Mean
Weak
Emotion
Mean
Strong
Emotion
%Bias t p
Log AnsLengthij
( )
5.58 5.44 -18.0 -31.56 0.00 5.44 5.44 0.1 0.13 0.90
Log AnsReadabilityij
( )
3.82 3.80 -5.2 -9.05 0.00 3.79 3.80 1.8 3.15 0.00
Log AnsSumj
( )
1.78 1.83 5.9 10.42 0.00 1.83 1.83 0.1 0.23 0.81
Log AnsReadij
( )
7.16 7.20 3.2 5.68 0.00 7.20 7.20 0.2 0.27 0.79
Log DifDaysij
( )
0.71 0.66 -4.3 -7.51 0.00 0.65 0.66 1.1 1.95 0.05
Rewardj0.36 0.38 4.5 7.95 0.00 0.39 0.38 -0.6 -0.98 0.33
QueContentWordj0.36 0.39 10.1 17.68 0.00 0.39 0.39 1.0 1.67 0.10
QueSentSumj6.45 6.48 0.7 1.19 0.24 6.48 6.48 -0.0 -0.08 0.94
TagSumj2.19 2.25 8.3 14.48 0.00 2.25 2.25 -0.0 -0.01 0.99
Volume 32 • Issue 1
34
Table 10. Differences in Mean Before and After CEM
Before CEM After CEM
L1
Mean Differences of Weak
Emotion and Strong
Emotion
L1
Mean Differences of
Weak Emotion and
Strong Emotion
Log AnsLengthij
( )
0.079 -0.138 0.050 0.002
Log AnsReadabilityij
( )
0.052 -0.025 0.058 0.001
Log AnsSumj
( )
0.036 0.045 0.008 -0.001
Log AnsReadij
( )
0.035 0.039 0.033 0.002
Log DifDaysij
( )
0.028 -0.050 0.004 0.000
Rewardj0.022 0.022 0.000 -0.000
QueContentWordj0.047 0.025 0.025 -0.000
QueSentSumj0.012 0.035 0.017 -0.033
TagSumj0.033 0.065 0.000 0.000
Figure 4. Standardized Percentage Bias for Each Covariate Before and After PSM
Volume 32 • Issue 1
35
Dongmei Han is a professor and the Doctoral Advisor at the School of Information Management and Engineering,
Shanghai University of Finance and Economics, China. She received her Ph.D. in Quantitative Economics from
Jilin University in 2002. Her research interests include finance engineering, management science and engineering.
She has published in the Information and Management and Expert Systems with Applications.
Lifeng He currently serves as an Assistant Professor at Qingdao University. He earned his Ph.D. in Management
from Shanghai University of Finance and Economics. His research interests include social media, online user
behavior and platform technology.
Wenfei Zhao is a student pursuing a Ph.D. degree at the school of Information Management and Engineering,
Shanghai University of Finance and Economics, Shanghai, China. Her research interests include information
system and business intelligence.
Dr.Xiaohang Zhou earned her Ph.D. in Management from Shanghai University of Finance and Economics. She
received a bachelor degree of Science and a master degree of Economics. She currently serves as an Assistant
Professor at Qingdao City University. Her research interests include online user behavior, social networks and
platform technology.
Figure 5. Propensity Score Histogram by Treatment Status
Available via license: CC BY 3.0
Content may be subject to copyright.
