Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis

DECSUP-12680; No of Pages 13 Decision Support Systems xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Decision Support Systems journal homepage: www.elsevier.com/locate/dss

Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis Mingguo Li a, Atreyi Kankanhalli b, Seung Hyun Kim c,⁎ a b c

Carnegie Mellon Unviersity, 5000 Forbes Ave, Pittsburg, PA 15213, USA School of Computing, National University of Singapore, 15 Computing Drive, Singapore 117418 School of Business, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 120-749, Republic of Korea

a r t i c l e

i n f o

Article history: Received 26 August 2014 Received in revised form 17 January 2016 Accepted 17 January 2016 Available online xxxx Keywords: User innovation community User ideas Message persuasion Cognitive overload

a b s t r a c t Online user innovation communities are increasingly being deployed by firms to garner innovation ideas from customers or users. However, very few ideas from such communities are successful in getting selected for implementation by the host firm. Given the limited understanding of the phenomenon, this study examines the determinants of firms' implementation of customers' ideas from user innovation communities. Drawing on theories of message persuasion and cognitive overload, we develop a conceptual model to explain how the likelihood of idea implementation is affected by the characteristics of its contributor as well as the characteristics of a submitted idea and its presentation. Specifically, we study the effects of the contributor's prior participation and prior implementation rate, as well as the idea's popularity, length, and supporting evidence on the idea's implementation likelihood. Our model is validated through logistic regression on a secondary dataset of 19,964 user ideas collected from two large user innovation websites, Salesforce.com IdeaExchange and Dell IdeaStorm. The results show significant impacts of these characteristics on idea implementation likelihood and also reveal important differences in their effects for hybrid (i.e., Dell IdeaStorm) versus professional (i.e., Salesforce.com IdeaExchange) user innovation communities. © 2016 Published by Elsevier B.V.

1. Introduction Innovation is a critical activity for sustaining firms' competitiveness in the market [7]. As a result, firms continue to invest in the development of new products, services, and processes. However, managers are concerned about how to encourage innovation while reducing its costs and risks. An approach to mitigate the risks and cost of innovation is to involve customers or users in the process [65,70]. For instance, in a study conducted at 3 M, innovations from users were found to generate more sales than traditional market research techniques [39]. By involving customers in the process of innovating, firms may benefit from lower development costs and enhanced customer acceptance of the innovations [64]. To formalize this approach, online user innovation communities are increasingly being deployed by firms to source for users' innovation ideas and preferences. As examples, Salesforce.com, Dell, and Starbucks have been pioneers in launching user innovation communities. By implementing ideas from its users, Dell introduced new options for its personal computers, such as installing Linux as an operating system [13]. Salesforce.com enhanced its customer relationship management (CRM) software by building new features adopted from ⁎ Corresponding author. Tel.: +82 2 2123 2506. E-mail addresses: [email protected] (M. Li), [email protected] (A. Kankanhalli), [email protected] (S.H. Kim).

its user innovation community, such as a mobile platform CRM. Starbucks introduced the customer idea of splash sticks poked into a hole on the top of its to-go cups to prevent the beverage from spilling out.1 Despite the potential value of sourcing innovation ideas from users, companies face challenges in setting up these communities, assessing a large number of submitted ideas, and obtaining valuable ideas from them [57]. Many firms do not have clear criteria to assess the submitted ideas and suffer from a lack of manpower and systematic processes to evaluate them [13,57]. At the same time, users also face challenges in getting their ideas implemented by host firms after investing their time and intellectual capital to generate them. With the typically low percentage of user ideas that are chosen for implementation,2 users would want to know how they could improve the likelihood of their ideas being selected. With prior research on online user innovation communities mainly focusing on identifying users' motivations for contributing ideas [4,23,24,35,37,38], there is limited study of the factors that influence firms' implementation of user ideas. Among the few studies in this area, Di Gangi and Wasko [14] employed the diffusion of innovations (DOI) theory to examine how the inherent characteristics of user ideas affect their implementation in Dell IdeaStorm. However, 1 2

http://blogs.starbucks.com/blogs/customer/archive/2008/04/09/splash-sticks.aspx. Dell indicates about 2.8% of user ideas are implemented, http://www.ideastorm.com/.

http://dx.doi.org/10.1016/j.dss.2016.01.004 0167-9236/© 2016 Published by Elsevier B.V.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

2

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

they found that idea characteristics such as relative advantage and compatibility were difficult for firms to predict and therefore did not influence idea implementation. Focusing on idea contributors, Bayus [4] and Huang et al. [29] used a learning perspective to explore how users' idea contribution behavior and their implementation rate change over time, rather than examining the implementation likelihood of individual ideas. Further, most prior studies used data from a single user innovation website, e.g., Dell IdeaStorm [4,14,29]. This could limit the generalizability of their findings, as the differences across communities, e.g., [7], are not considered. Overall, our review indicates a lack of understanding of the factors influencing the implementation likelihood of an idea, and that too across different user innovation communities. The practical and theoretical issues mentioned above motivate us to holistically (by including both idea and contributor characteristics) examine the antecedents of idea implementation likelihood in user innovation communities by employing alternative theories that could explain the phenomenon. In reality, a user innovation community is typically characterized by low review capacity of the firm, i.e., there is insufficient manpower to review the numerous user ideas in detail when a large number of ideas are submitted every day [11,57]. In such a context, we propose that user ideas will be selected for implementation if they are persuasive and, at the same time, their presentation does not cognitively overload readers (i.e., both firm's reviewers and other community members). Accordingly, we develop a model based on theories of persuasion [46,49–51] and cognitive overload [32] to explain the likelihood of user idea implementation. Through the model, we aim to answer three fundamental questions: (1) What user/contributor characteristics influence the implementation likelihood of their ideas by firms? (2) What characteristics of the idea and its presentation influence their implementation likelihood by firms? and (3) How do the effects of idea presentation characteristics differ across the type of user innovation community (i.e., professional communities with corporate members vs. hybrid communities with both corporate and individual members)? The model is tested with secondary data on 19,964 user ideas collected from two large user innovation communities, Salesforce. com IdeaExchange (a professional community) and Dell IdeaStorm (a hybrid community). In terms of theoretical contributions, this study is novel in (1) examining the factors leading to the implementation of user ideas based on persuasion and cognitive overload perspectives, (2) considering both contributor and idea (including presentation) factors as antecedents, and (3) comparing the differential effects of idea presentation factors across two types of user innovation communities. Further, by answering the above questions, this study suggests a number of practical implications for management. For firms that are aiming to launch online user innovation communities, our findings from these successful communities can provide guidelines on what kind of user ideas are being implemented, how to filter ideas and assess them, especially when there are a large number of ideas and limited resources or capacity to process them. It can also help firms to identify the contributors who may potentially submit valuable ideas and respond to their ideas quickly in order to incentivize them. Based on our findings about the characteristics of implemented ideas and their presentation, online user innovation communities can provide their members with guidelines on how to position and present their ideas. For users, adopting these guidelines could help them draw more attention to their ideas from firms' reviewers and other members, than those contributors not following the guidelines. Last, based on the differences in effects of idea presentation characteristics found across the two types of user innovation communities, the guidelines for idea presentation could be modified for each type of community. 2. Conceptual background We first review the related studies on user innovation communities to indicate the gap in the literature that our study seeks to address. We

then introduce the message persuasion perspective that helps us to identify contributor, idea, and presentation factors that make the idea posting persuasive. We subsequently apply cognitive overload concepts to explain the relation between idea's presentation characteristics and its implementation likelihood. 2.1. User innovation communities Innovation is a process wherein firms transform ideas into new or improved products, services, or processes [3]. While in the past innovation ideas were thought to originate from organizations alone, it is now clear that users can play a key role in innovation [8,65]. Prior research on user innovation has focused on two related issues, the motivations of users to innovate [21–24,36–38,58,62] and how to support and engage users as innovators [16,37]. Various ways have been employed to engage customers/users for innovation, such as providing them with toolkits to create their own innovations [66], talking to lead users during the innovation process [39], providing virtual customer environments [45], running online contests [69], and establishing brand communities for contribution of user innovation ideas [23]. Here, we focus on user innovation communities that are increasingly receiving attention from researchers and practitioners as a means of garnering customer ideas [68]. Members of user innovation communities are valuable sources of innovation because of their passion, experience, and cooperation in knowledge generation [23]. Nonetheless, little research has been conducted to understand the factors that influence the implementation of user ideas in such communities. Of the limited studies on this topic, Di Gangi and Wasko [14] used DOI theory to examine the determinants of idea implementation by analyzing 21 ideas submitted to Dell IdeaStorm. They found that the idea characteristics of relative advantage and compatibility did not influence idea implementation and suggested that this is because these characteristics were difficult for firms to predict. Rather, idea age and complexity were found to affect idea implementation, while idea popularity did not. However, the study sample was small and some results were obtained by examining few, e.g., 2 ideas. In another study in the same community, Bayus [4] focused on idea contributors, rather than ideas. He found that past success of the contributor is negatively related, while diversity of their past commenting activity is positively related to their idea implementation likelihood. Finally, Huang et al. [29] also examined idea contributors in Dell IdeaStorm and observed that they learn how to come up with high-potential ideas over time through participation and peer voting on ideas. Contributors of low-potential ideas eventually become inactive, while contributors of high-potential ideas remain active, which somewhat contradicts the results from Bayus [4] that there is a negative effect of prior success on idea implementation. As per Huang et al. [29], over time, the average potential of generated ideas increases, while the number of ideas contributed decreases. While these studies have enhanced our understanding of the phenomenon, we identified several gaps in the prior literature based on our review. First, the prior empirical studies implicitly assume that the idea submissions are fully assessed by the firm's reviewers. However, this overlooks the limited firm resources as well as “bounded rationality” [61] of reviewers. Firms' review capacity in user innovation communities is typically low as a large number of ideas are submitted every day with insufficient manpower to process them in detail [11,57]. For example, it is estimated that it cost “approximately $500 and took four hours of staff and management time to process each idea” even in a conventional suggestion system within a firm [52]. As a user innovation community usually covers a fairly wide range of ideas from a diverse set of contributors, the reviewers from the firm are unlikely to be familiar with all the idea topics. The difficulty of processing many ideas was also mentioned by a Dell IdeaStorm user: “Now they have more people working on the site but the duplicates and backlog of work to catch up on is causing the delay in response to continue” [15]. With the large number of ideas typically posted, reviewers (and other community members) look for fast

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

ways to filter ideas. Thus, we argue that the persuasiveness of an idea and its presentation become vital to its success in such a context. Second, while prior studies have explored a few characteristics of ideas and contributors, the effects of idea presentation factors have not been examined. Indeed, managers and users would be interested to know how user ideas should be presented to draw attention and be selected for implementation. Thus, we extend prior research by examining the combination of contributor and idea characteristics, and adding idea presentation factors to enhance the explanation of idea implementation in these communities. Third, prior studies on user innovation communities have mainly focused on one website, e.g., Dell IdeaStorm [4,15,29]. This limits the generalizability of findings as the heterogeneity across different types of communities is not considered. The above gaps motivate us to holistically examine the antecedents of idea implementation using data collected from two user innovation communities of different types. 2.2. Message persuasion As user innovation ideas are contributed in the form of messages or postings in online communities, we propose that the persuasiveness of the message influences its likelihood of implementation. In this regard, persuasion theories can explain what factors make a message persuasive and thereby influence its adoption and implementation [9,63]. From ancient times, philosophers like Aristotle have studied the modes of persuasion, where persuasion is defined as making the recipient accept a point of view and/or undertake a course of action [46]. Specifically, Aristotle suggested three modes of persuasion [2]: ethos (appeal to authority or credibility of the source), pathos (appeal to audience's emotions), and logos (appeal to logic). In the context of user innovation communities, the ethos and logos modes are likely to be most relevant when contributors need to persuade reviewers and other members about the merit of their ideas. More recently, cognitive dissonance [19], social judgment [59], and message persuasion [50] theories have been used to explain how persuasion takes place. Cognitive dissonance theory posits that people experience dissonance when they have contradictory thoughts or ideas about something [19]. Dissonance is an unpleasant state that encourages attitude change to achieve or restore consonance or consistency. Thus, people can be persuaded to change their attitude by creating dissonance in their mind about the subject. Social judgment theory modifies this argument by proposing that people judge messages by how much the message agrees or disagrees with their own attitude [59]. People can react differently to a message depending on their existing attitude or anchor. While cognitive dissonance and social judgment theories focus on modifying the existing attitude of the message recipient, e.g., convincing a smoker not to smoke, this is less relevant for our study where the reader (reviewer or community member) of the idea posting may not have an existing attitude about the idea which needs to be changed. But a message persuasion perspective which proposes various characteristics of the message and source that determine its persuasiveness [50] appears more relevant to our study. As per this perspective, message recipients will vary in the extent to which they cognitively process a particular message, which in turn impacts the success of the persuasion attempt [49,50]. When the recipient is able and willing to cognitively elaborate on a message, the quality of the arguments contained within the message will determine its degree of influence on him or her [5,50,63]. The argument quality of ideas will be high if the source (i.e., user contributor in our case) presents the ideas based on reasoned arguments [26] – akin to Aristotle's logos mode. Indeed, the quality could be enhanced if the contributor thoughtfully formulates and articulates the innovation idea and provides supporting evidence for it [25]. Alternatively, when recipients do not have the ability or willingness to process messages in detail (as in the case of user innovation communities with low review capacity), they will still look for cues of message quality [50,51].

3

As the theory suggests, we propose that the firms' reviewers (who are unable to process the large number of innovation ideas in detail) will look for cues in the form of supporting evidence as indicators of the argument quality of the user idea [50]. Other than supporting evidence, we posit that there are other cues related to the user idea itself as well as its source or contributor that determine its persuasiveness and thereby its implementation likelihood. For example, product popularity signaled by the number of online reviews has been considered as an important indicator for persuading people to purchase products [47,48]. Analogous to product popularity, we expect idea popularity (e.g., indicated by the number of likes) to influence firms' reviewers to select a user innovation idea for implementation. Additionally, recipients often use cues pertaining to the message's source as indicators of quality when they are unable or unwilling to expend the effort to elaborate on the message's content [50] – akin to Aristotle's ethos mode. Specifically, the source's credibility, i.e., the extent to which the source is perceived to be knowledgeable, competent, and trustworthy [5], can act as a cue for recipients to adopt a new piece of knowledge [5,63]. As per the message persuasion theory, the source's credibility, signaled by the prior experience and prior performance of the source, acts as a cue to make the message more persuasive [5,50]. In the context of our study, the credibility of the idea contributor is indicated by their prior participation and prior implementation rate, which could thereby influence the idea's persuasiveness. Other than the contributor (prior participation, prior implementation rate), idea (popularity), and presentation (supporting evidence) antecedents identified above, we also posit that the idea presentation should not cognitively overload reviewers (who need to select the idea) and other members (who vote on or like the idea) if it is to be selected among many submitted ideas. 2.3. Cognitive overload Cognitive overload, which is nowadays often equated with information overload [34], is defined as the state of an individual (or system) in which too many inputs do not allow the inputs to be processed and utilized [55]. Indeed, psychologists have long recognized the limited capacity of people to store current information in memory. Particularly influential in this regard was Miller's idea that humans can process seven chunks of information [44]. More broadly, people have finite limits to the amount of information they can assimilate and process at one time [44]. When these limits are exceeded, “overload” results whereby decision making is hampered. Subsequent research has gone on to study the causes and consequences of overload in organizational contexts e.g., [34] as well as in the society at large [55], especially with the proliferation of Internet technology. In the context of online communication, it has been suggested that people may be overloaded when (1) they receive too many messages [67] and (2) cannot recognize the significance of incoming messages when the messages are not precisely organized [28]. In our study context, the first condition applies as reviewers and members in online user innovation communities are unable to process user ideas in detail when a large number of ideas are submitted every day [11,57]. We are thus interested to understand the characteristics which make certain user idea messages be recognized (as a counterpoint to the second condition) and implemented even in such conditions of low reviewer capacity. In this regard, prior research found that members in online communities are more likely to respond to shorter messages than to long ones [32]. In a similar vein, other work [27] suggests that longer messages are likely to decrease message clarity and argument quality and make the message challenging to process. Thus, we include idea length as an idea presentation characteristic that may influence its persuasiveness and implementation likelihood. Further, we proposed earlier that when reviewers do not have the ability or willingness to process the innovation idea in detail, they may look for cues of the supporting evidence (e.g., reference pages and

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

4

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

supplementary images in the message) to indicate its argument quality and determine its implementation likelihood. However, here too, the principles of cognitive overload are likely to operate. To start with, having relevant reference pages and supplementary images could convince reviewers of the quality of the user innovation idea. Yet, when the number of reference pages and supplementary images exceeds a certain threshold, as per the theory [10],3 overload can set in whereby the message may not be evaluated appropriately and adopted. Thus, we use cognitive overload theory to explain the relationships between the idea presentation characteristics (length and supporting evidence) and implementation likelihood in our model. 3. Model and hypotheses As described above, we derive the antecedents of idea implementation likelihood in a user innovation community based on the message persuasion and cognitive overload perspectives. Under conditions of low review capacity in such contexts, we propose that the implementation likelihood of a submitted idea will be affected by its contributor, idea, and presentation characteristics. These characteristics identified in the previous section are observable by reviewers and other members in the user innovation community and can determine how persuasive the idea is in order to be selected for implementation. Indeed, user innovation communities like Dell IdeaStorm and Salesforce.com IdeaExchange compile and display contributor and idea information. These include idea contributors' prior participation, prior implementation rate, and the idea's popularity for each idea submitted. Further, as discussed in the previous section, the firm's reviewers can use cues of idea length and supporting evidence to gauge the argument quality of the submitted idea. In addition, we propose that the way innovation ideas are presented and their effects vary across different types of user innovation communities. Here, we examine how the distinct context of such communities (i.e., professional vs. hybrid communities) may moderate the effects of idea presentation characteristics on their implementation likelihood. Our research model is shown in Fig. 1, where the unit of analysis is a submitted idea and the DV is its implementation likelihood (our results are also shown in the figure, which will be explained later). 3.1. Source/contributor characteristics 3.1.1. Prior participation In a user innovation community, members have distinct participation histories in terms of commenting on other members' ideas. The participation of users by commenting can be viewed as a process of informal learning about the brand and its products. Informal learning refers to the acquisition of knowledge which occurs without the presence of externally imposed (e.g., curricular) criteria [40]. Participation in the community enhances individual's knowledge of the firm's values, market orientation, and present needs [11] by repeated interactions with community moderators and other members. Such knowledge can be transformed into a greater level of relevance and practicality of the participant's contributions. Consequently, an innovation idea from a user with higher prior participation may be more valuable to the company and adopted by it. This reasoning follows the persuasion perspective where prior experience of the source has been suggested as a signal of source credibility that determines the message's influence [5,50]. The firm's reviewers could perceive contributors' previous participation as an indicator of their credibility and the idea's attractiveness. This is 3 Clevinger (2014) notes that when task difficulty (in this case, comprehending the number of supporting evidences) increases upto a certain point, arousal (stimulation) increases and helps performance (evaluating the idea). After that as the difficulty (number of supporting evidences) increases further, the increasing stimulation overloads individual's memory and becomes detrimental to performance, whereby the innovation idea could not be evaluated appropriately.

likely when the reviewers are constrained to evaluate the idea content (as in the user innovation community contexts described above) and look for indicators to guide their idea selection. Thus, H1a. Prior participation is positively related to the idea implementation likelihood.

3.1.2. Prior implementation rate Prior implementation rate refers to the rate of successful implementation of the contributor's previous ideas in the user innovation community. The prior implementation rate would thus vary across contributors and over time for a contributor. The prior implementation rate discloses information about several aspects of the contributor. A contributor with a higher prior implementation rate is expected to be more knowledgeable about the brand and its products. Accordingly, such a contributor is likely to possess greater capability to develop valuable and relevant innovation ideas for the firm. Innovation ideas from a contributor with higher prior implementation rate are thus expected to be of higher relative value and relevance to the firm. Such logic follows the persuasion perspective [5,50] where the firm's reviewers may consider the prior implementation rate of the contributor as a signal of the credibility of the contributor and attractiveness of the idea. Thus, H1b. Prior implementation rate is positively related to the idea implementation likelihood.

3.2. Idea characteristics 3.2.1. Idea popularity A user innovation community of a firm or brand typically consists of customers enthusiastic about the particular brand [20]. In most such communities, members can indicate their preferences about an innovation idea by “promoting” or “demoting” the idea. As an innovation idea will be promoted when it is preferred by other members, an idea with a high voting score is considered popular in the community. The voted popularity of a prospective product idea is often an indicator of its potential acceptance and popularity in the market [56]. Thus, its popularity in the user innovation community can indicate to the firm's reviewers the potential market success of the innovation idea. A user innovation community as a test market is particularly useful for a firm when it is difficult to predict customers' reactions to the innovations. Idea popularity acts as a signal for future acceptance of the implemented idea and can draw a favorable response from reviewers. This reasoning agrees with the persuasion perspective where idea popularity is seen as an indicator of its value or quality. Thus, H2. Idea popularity is positively related to the idea implementation likelihood.

3.3. Idea presentation characteristics 3.3.1. Idea length When a user contributes ideas in a user innovation community, the length of the message posting can determine how easy it is to comprehend. Linguistics research indicates that longer messages, usually entailing complex structures, are more difficult to understand and are less persuasive [41]. In general, longer messages are likely to decrease message clarity and argument quality [27]. Similarly, the literature on cognitive overload suggests that message length is an indicator of message complexity and determines the likelihood that users will respond to the message under information overload [32]. In the persuasion literature, too, the length of the message has been found to negatively affect its quality [30,31]. Given the limited review capacity of an online user innovation community to process all submitted ideas, reviewers may gauge that lengthy ideas are too complex to understand and implement

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

5

Fig. 1. Research model.

and may not select them. Indeed, innovation communities such as Dell IdeaStorm have emphasized the importance of summarizing the idea to catch readers' attention.4 A lengthy description may even reflect users' inability to present their ideas concisely and clearly. Hence, H3a. Idea length is negatively related to the idea implementation likelihood.

3.3.2. Supporting evidence When a member contributes an idea in a user innovation community, the contributor can add references to support the idea description by inserting hyperlinks to other web pages on the Internet. Including relevant reference pages in the idea description is likely to enhance the quality of the message for several reasons. First, adding reference pages to the idea description should improve the strength of the explanation since the information presented on a referenced webpage is often more rigorous and professionally written than the description produced by the user contributor. In fact, a lack of idea justification is identified as one of the main challenges faced by user innovation communities [15]. Second, referenced pages are mostly selected from credible information sources such as online articles written by journalists or the brand's own web pages (as confirmed by our qualitative analysis of contributed ideas in the two communities in our sample). For example, contributors in Dell IdeaStorm often point to Dell's own product support web pages to highlight a current problem and the inconvenience caused by it. Including references can thus enhance the argument quality and persuasiveness of the innovation idea. Third, it is difficult for users in the community to communicate well through the limited media when an idea involves a tacit element of experience [15]. Here, too, reference pages can help to communicate the idea more lucidly. While the presence of reference pages in the idea description should increase its persuasiveness, this would be valuable up to a certain extent. If a large number of reference pages are used to provide evidence for the idea, it is likely that reviewers would be overwhelmed and it 4 The Dell IdeaStorm website guide for contributors says “Once you have an interesting idea, it is important to have a compelling title and a good summary.”

would detract from the evaluation of the idea as suggested by the cognitive overload literature [34,54]. For reviewers in the user innovation community, a large number of reference pages would lead to the situation of information overload whereby the idea may be difficult to process.5 Following this line of reasoning, an inverted U relationship is expected between the number of reference pages of an idea and its implementation likelihood. Thus, we hypothesize. H3b. Number of reference pages has an inverted U relationship with the idea implementation likelihood. In user innovation communities, a contributor can also use images in his/her idea description. By including images in the idea description, the contributor improves the media richness, i.e., the ability to change understanding within a time interval, of the message [12,60]. Media richness has been associated with message persuasion due to higher levels of comprehension and vividness [1]. In the context of our study, higher media richness of the idea posting through use of images can make the idea more concrete and understandable to the reviewer while assessing the idea. Furthermore, an idea posting with images is likely to draw more attention from the community's reviewers. For instance, a contributor in Dell IdeaStorm posted an image of a motherboard design to argue that a visual image is easier to understand and more helpful for self-repair and user maintenance than describing similar contents in text in an electronic users' manuals with many pages. However, as with supporting evidence through reference pages, the use of supplementary images would be useful up to a certain extent. If there are many supporting images, it may also be cognitively demanding to the reviewer who is processing this idea. The resultant cognitive overload can deter selection of the idea. This agrees with the cognitive overload literature [10] that while moderate mental stimulation is beneficial, too much stimulation impairs task performance. Hence, we hypothesize.

5 Although the reference pages are not embedded in the text, the reviewers will most likely look up at least some of the referenced pages once they are provided together with the main idea. Browsing through many reference pages will overload the reviewers and hamper processing of the main idea.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

6

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

H3c. Number of supplementary images has an inverted U relationship with the idea implementation likelihood. 3.4. Type of user innovation community We propose that the above relationships (H3a, H3b, and H3c) between idea presentation characteristics and their implementation likelihood are moderated by the nature of members and their communications in the user innovation community. As different types of communities differ in the nature of ideas submitted and their presentation, the effects of idea presentation characteristics would differ across the communities. Particularly, members of a professional user innovation community such as Salesforce.com IdeaExchange are typically employees or representatives of a client firm with professional background, who come up with innovation ideas that specifically address their current task requirements [6] and present them precisely. For example, users in Salesforce.com IdeaExchange often submit ideas that are very precise and specific, e.g., “Currently, it is not possible to use a picklist in a formula field just by inserting the merge field. Creating a text formula field, one would think that it is possible to display a value of picklist just by inserting the merge field into the formula. However, without using the ISPICKVAL function a picklist cannot be used in a formula field.”6 In contrast, non-professional communities with individual members or hybrid communities with both individual and corporate members would, in general, see less specific and thought-out ideas that are more casually presented. For example, in Dell IdeaStorm, users often post general ideas without considering the firm's capability or idea feasibility. For one of the popular ideas titled “Provide Linux Drivers for all your Hardware,” Dell responded, “Drivers are available for Linux to make some but not all hardware work. We continue to work with our partners to provide for better hardware support.”7 Thus, members in professional communities are more likely to communicate professionally and hence more capable of describing their innovation ideas precisely than users in non-professional or hybrid settings. Indeed, professional communication is typically more precise than non-professional communication [18]. We earlier mentioned that, based on the literature on cognitive overload [32] and persuasion [30,31], message length is an indicator of message complexity and negatively affects message quality. Given more professional communications, reviewers in professional user innovation communities are likely to perceive that the complexity does not increase as much merely because of idea length (as the description is still precise). Thus we expect a lesser negative impact of idea length on implementation likelihood in professional than in hybrid communities. H4a. The negative effect of idea length on idea implementation likelihood is lower in a professional user innovation community than in a hybrid community. A similar line of reasoning suggests a moderating effect of the type of user innovation community on the relationship between supporting evidence and the likelihood of idea implementation. Members in a professional user innovation community may, in general, be more capable of taking advantage of reference pages and supplementary images to support their ideas in a more convincing manner. In contrast, in nonprofessional or hybrid communities, reference pages and supplementary images inserted by less professional members may increase the cognitive overload among reviewers as they may be less relevant for supporting the idea. Thus, reviewers are likely to respond to an increase in supporting evidence more positively in a professional community

6 7

http://success.salesforce.com/ideaView?id=08730000000Brl6 http://www.ideastorm.com/ideaView?id=0877000000006wRAAQ

Table 1 Description of variables.

Idea implemented IdeaStorm IdeaExchange Prior implementation rate IdeaStorm IdeaExchange Prior participation IdeaStorm IdeaExchange Idea popularity IdeaStorm IdeaExchange Idea length IdeaStorm IdeaExchange Number of reference pages IdeaStorm IdeaExchange Number of supplementary images IdeaStorm IdeaExchange Age of community IdeaStorm IdeaExchange Tenure in community IdeaStorm IdeaExchange Same day submission IdeaStorm IdeaExchange IdeaStorm category 1 IdeaStorm category 2 IdeaStorm category 3 IdeaExchange category 1 IdeaExchange category 2 IdeaExchange category 3 IdeaExchange category 4 IdeaExchange category 5

N

Mean

Std. Dev.

Min

Max

19,964 9984 9980 19,964 9984 9980 19,964 9984 9980 19,964 9984 9980 19,964 9984 9980 19,964 9984 9980

0.03 0.17 0.00 1.00 0.02 0.15 0.00 1.00 0.04 0.19 0.00 1.00 0.01 0.08 0.00 1.00 0.01 0.05 0.00 1.00 0.02 0.10 0.00 1.00 63.37 285.87 0.00 2966.00 118.32 395.53 0.00 2966.00 8.40 30.68 0.00 580.00 353.40 2166.17 −1460.00 118,080.00 427.23 2,868.10 −1,460.00 118,080.00 279.54 1,070.86 −120.00 37,110.00 93.17 94.21 1.00 2,502.00 112.17 117.22 1.00 2,502.00 74.17 57.32 1.00 1,542.00 0.13 0.86 0.00 61.00 0.23 1.08 0.00 61.00 0.03 0.52 0.00 47.00

19,964

0.12

0.40

0.00

11.00

9984 9980 19,964 9984 9980 19,964 9984 9980 19,964 9984 9980 9984 9984 9984 9980 9980 9980 9980 9980

0.05 0.18 21.87 14.00 29.76 5.61 3.87 7.36 20.84 28.06 13.62 0.61 0.34 0.04 0.17 0.20 0.54 0.01 0.02

0.35 0.43 16.13 13.18 14.93 8.76 6.58 10.20 33.16 44.91 8.79 0.44 0.43 0.17 0.24 0.22 0.30 0.09 0.11

0.00 0.00 0.00 0.00 0.00 1.00 1.00 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

11.00 9.00 48.00 44.00 48.00 48.00 45.00 48.00 203.00 203.00 41.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

than in a hybrid community if the number of supporting evidence is reasonable. But, when the number of supporting evidence is high and information overload is present, an increase in supporting evidence will be perceived more negatively in a hybrid community than in a professional community.8 That is, reviewers will perceive more cognitive overload by excessive supporting evidence and respond more sharply to it in a hybrid community than in a professional one. Thus, we propose, H4b (H4c). The relationship between the number of reference pages (supplementary images) and the idea implementation likelihood is moderated such that the positive relationship between the two is greater, while the negative relationship between the two is lower in a professional community than in a hybrid community.

4. Research method 4.1. Data collection We chose two online user innovation communities, i.e., Salesforce. com IdeaExchange and Dell IdeaStorm for our study, which represent 8 It mathematically means that the first order derivative (or the slope) is larger at any level of supporting evidence in a professional community than in a hybrid community. For example, suppose there is a function of x, f(x) = ax2 + bx + c where x denotes the level of supporting evidence. Then, its first order derivative is dy/dx = f’(x) = 2ax + b. We hypothesize b(professional community) N b(hybrid community) such that for any level of supporting evidence, f’(professional community)(x) N f’(hybrid community) (x) because 2ax + b(professional community) N 2ax + b(hybrid community).

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

7

Table 2 Correlations of variables.

(1) Idea implemented (2) Prior participation (3) Idea popularity (4) Prior implementation rate (5) Message length (6) Number of reference pages (7) Number of supplementary Images (8) Tenure in community (9) Age of community (10) Same day submission

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

0.04*** 0.09*** 0.07*** −0.02*** 0.02*** 0.05*** 0.02*** −0.07*** −0.02***

0.00 0.07*** 0.03*** 0.12*** 0.05*** 0.29*** −0.05*** −0.07***

0.01 0.02** 0.10*** 0.02*** −0.03*** −0.11*** 0.09***

−0.01 0.00 0.09*** 0.22*** 0.02*** −0.06***

0.25*** 0.03*** −0.03*** −0.08*** 0.02***

0.10*** 0.03*** −0.07*** 0.00

0.16*** 0.06*** −0.07***

0.31*** −0.13***

−0.30***

*** Significant at 1 %, ** Significant at 5 %, * Significant at 10 %.

a professional community and a hybrid community, respectively. The two communities were chosen for their popularity and the publicly available data on the activities of members and the adopting firms in the communities. Additionally, both communities run on the same platform since Dell adopted the platform solution provided by Salesforce. com. That is, both communities have similar user interfaces and procedures for reviewing and implementing innovation ideas. Furthermore, the two communities were chosen for our study since they were launched around the same time. These similarities between the two communities help to ensure that differences found between the two communities are not attributable to website-level differences such as the user interface and user experience, but to their different membership nature (i.e., professional vs. hybrid community). Salesforce.com specializes in enterprise software solutions and is best known for its on-demand customer relationship management (CRM) products. Salesforce.com IdeaExchange was launched to collect innovation ideas from its clients in September 2006. Dell is one of the leading global providers of computer products and solutions. Dell IdeaStorm was established “as a way to talk directly to customers” in February 2007. In both communities, members can contribute their innovation ideas after registration. They can also make comments on any posted idea and vote on the ideas in the website. When submitting, an idea can be placed into several categories by a contributor. Members can view the submitted ideas by category or by their implementation status. Various features such as greeting other members after logging in, the statistics of members' past activities, and their profile information are available for others to view to facilitate information sharing in the communities. The two communities do not provide monetary rewards for members' participation or contribution of ideas. Since Salesforce.com produces business software and most contributors of ideas in its community are users of its CRM products in their workplace, Salesforce.com IdeaExchange can be considered as a professional user innovation community. In contrast, many idea contributors on Dell IdeaStorm are individual consumers of Dell's personal computer products. Thus, Dell IdeaStorm can be considered as a hybrid user innovation community with both individual and professional consumers. The ideas were collected from the two user innovation communities right from the establishment of the websites, across a period of 48 months for Salesforce.com IdeaExchange and 44 months for Dell IdeaStorm. Our dataset consists of 9980 innovation ideas from Salesforce.com IdeaExchange and 9984 innovation ideas from Dell IdeaStorm. Among these, 221 ideas (2.21% of total) from Dell IdeaStorm and 381 ideas (3.82% of total) from Salesforce.com IdeaExchange have been implemented. 4.2. Measurement of variables The unit of analysis in our model is an innovation idea and the dependent variable is the implementation status of the idea. In both

communities studied, the status of the contributed idea is exhibited next to each idea. We coded the dependent variable as 1 if it has been implemented, and 0, otherwise. For the source and idea characteristics, idea contributor's prior participation was measured by the number of comments the contributor of the idea has made before the current idea. Commenting is one of the most frequent activities performed by online user innovation community members and is highly correlated with their prior contribution of ideas (correlation of 0.88). Idea contributor's prior implementation rate was calculated as the total number of implemented ideas divided by the total number of contributed ideas before the user's current idea contribution. For a first time contributor, his/her prior implementation rate was considered as 0.9 In the communities studied, a member can vote on an innovation idea by “promoting” or “demoting” it. In both these communities, the policy is to augment (deduct) the voting score by 10 points if the idea is promoted (demoted) by a member. The total voting score of an idea was used to measure idea popularity. For the idea presentation characteristics, the idea length was measured by the number of words contained in the idea posting. It is to be noted that the communities do not specify any threshold for idea length. The supplementary image and reference page variables were measured as the number of supplementary images or reference pages that the idea description contains, respectively. On Salesforce IdeaExchange, 17% of innovation ideas contain at least one image, while only 3.7% of ideas on Dell IdeaStorm include images. In contrast, 13.1% of innovation ideas on Dell IdeaStorm contain hyperlinks to other websites, compared to 1.6% of ideas on Salesforce IdeaExchange in our dataset. To test the moderating effect of the community type, we used the dummy variable Professional Community with value 1 if the idea is from Salesforce.com, and 0, otherwise. We controlled for temporal characteristics such as the number of ideas submitted on the same day, the tenure of the idea contributor in the community, and the age of the community. Same Day Submission measures a possible constraint in resources in the communities in reviewing each idea thoroughly when a large number of ideas are submitted in a single day [11,57]. Tenure in Community was measured by the number of months elapsed since the idea contributor made his/ her first comments or contribution. Age of community is the number of months elapsed since the launch of the community at the time of the particular idea submission.

9 From the viewpoint of a reviewer or other members, no successful submissions from a contributor will not be impressive at all, and thus it is reasonable to treat it as a zero percent implementation rate. Empirically, we included another control variable to indicate whether an idea is a contributor's first submission, but the effect of this variable was insignificant. Therefore, we conclude that a discontinuity of this variable due to our coding has a minimal effect.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

8

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

Table 3 Estimation results. Variable

(1) Controls only

(2) Direct effects

(3)Direct & quadratic

(4)Direct + Quadratic + Interactions

Intercept Age of community Tenure in community Same day submission Prior participation Prior implementation rate Idea popularity Idea length Number of reference pages Number of supplementary images Number of reference pages2 Number of supplementary images2 Idea Length × Professional Community Number of References × Professional Community Number of Images × Professional Community Pseudo R-squared

−4.159⁎⁎⁎ (0.581) −0.048⁎⁎⁎ (0.004) 0.035⁎⁎⁎ (0.005) −0.256 ⁎⁎ (0.058)

−4.125⁎⁎⁎ (0.581) −0.042⁎⁎⁎ (0.004) 0.015⁎⁎ (0.006) −0.290⁎⁎⁎ (0.065) 0.160⁎⁎⁎ (0.031) 0.119⁎⁎⁎ (0.018) 0.222⁎⁎⁎ (0.050) −0.102⁎ (0.053) 0.051⁎⁎⁎ (0.016) 0.091⁎⁎⁎ (0.023)

−4.027⁎⁎⁎ (0.582) −0.040⁎⁎⁎ (0.004) 0.012⁎ (0.006) −0.269⁎⁎⁎ (0.067) 0.107⁎⁎⁎ (0.036) 0.123⁎⁎⁎ (0.019) 0.229⁎⁎⁎ (0.043) −0.115⁎⁎ (0.055) 1.948⁎⁎⁎ (0.355) 0.232⁎⁎⁎ (0.063) −0.642⁎⁎⁎ (0.221) −0.036⁎⁎ (0.016)

9.77%

13.23%

16.08%

−3.985⁎⁎⁎ (0.586) −0.042⁎⁎⁎ (0.004) 0.013⁎⁎ (0.006) −0.292⁎⁎⁎ (0.068) 0.107⁎⁎ (0.041) 0.114⁎⁎⁎ (0.020) 0.235⁎⁎⁎ (0.045) −0.431⁎⁎⁎ (0.126) 1.896⁎⁎⁎ (0.357) 0.215⁎ (0.119) −0.722⁎⁎⁎ (0.228) −0.031⁎ (0.018) 0.423⁎⁎⁎ (0.137) 0.703⁎⁎⁎ (0.184) 0.006 (0.111) 16.68%

Robust standard errors in parentheses. The coefficients for category variables are not shown for brevity. ⁎⁎⁎ Significant at 1%. ⁎⁎ Significant at 5%. ⁎ Significant at 10%.

In addition, we controlled for the heterogeneity of implementation likelihood across different categories of innovation ideas. When a user contributes an innovation idea, he/she can choose to place the idea under several categories. There were 82 subcategories in Salesforce. com IdeaExchange and 42 subcategories in Dell IdeaStorm at the time of data collection. These subcategories were grouped under five general categories in Salesforce.com IdeaExchange and three general categories in Dell IdeaStorm.10 We added dummy variables for these general categories in our model. The Category Dummies capture the differences in the difficulty of implementing an innovation idea across categories. For example, an innovation idea on the marketing strategy of the firm is typically much more difficult to implement than one on its website design. Since the categories are unique to each community, they capture the effects of community-level differences in addition to categoryspecific effects. Such community-level differences include culture and policies that may affect the idea implementation likelihood. Contributors may choose multiple categories or even not select any category for their contributions. We did not add dummies for subcategories to avoid severe multicollinearity. The descriptive statistics of the variables used in our empirical model are given in Table 1. Table 2 shows the correlations between model variables, indicating that all values are below 0.31.

4.3. Empirical model We use logistic regression to test our hypotheses with a binary dependent variable. Logistic regression has been employed to explain choice decisions of individuals or companies in various contexts [42,43]. We propose that a firm's decision to implement an innovation idea is determined by two source/contributor characteristics (prior participation and prior implementation rate of idea contributors), an idea characteristic (idea popularity), three idea presentation characteristics (idea length and supporting evidence, i.e., reference pages and supplementary images), the control variables, and an unobserved constant.

10 The five general categories for Salesforce.com IdeaExchange are “large enterprise,” “applications,” “force.com platform,” “appExchange,” and “non-product ideas.” The three general categories for Dell IdeaStorm are “product ideas”, “Dell ideas” and “topic ideas.”

The probability density of an innovation idea i to be implemented by the firm can be written as PrðImplementedi ¼ 1jXi Þ 0 1 α þ β 1  ðAge of Communityi Þ þ β2  ðTenure in Communityi Þ B þ β  ðSame Day Submission Þ C i 3 B C B C B þ β4  ðPrior Participationi Þ C B C B þβ5  ðPrior Implementation Ratei Þ þ β6  ðIdea Popularityi ÞC B C B C B þβ7  ðIdea Lengthi Þ þ β 8  ðReference Pagei Þ C B C B þβ  ðSupplementary Imagei Þ C 9 ¼ ΛB C B 2 2 C B þβ 10  ðReference Pagei Þ þ β 11  ðSupplementary Imagei Þ C B C B þβ  ðIdea Length Þ  ðProfessional Community Þ C i 12 i B C B C B þβ 13  ðReference Pagei Þ  ðProfessional Communityi Þ C B C B þβ  ðSupplementary Image Þ  ðProfessional Community Þ C i @ A 14 i X8   γ  Category Dummiesji þ ε i þ j¼1 j

where Λ(x) =ex/(1+ ex). α is a constant term; εi is the error term; βj can be interpreted as the change of likelihood of implementation made when each variable changes. More specifically, βj in logistic regression describes the size of contributions of independent variables to the log of odds ratio which is defined as the ratio of the probability that an event would occur to the probability that an event would fail to occur (i.e., Pr(Implementedi =1)/Pr(Implementedi = 0).11 Age of Community, Tenure in Community, Same Day Submission, and Category Dummies are the control variables in our study. Notably, Professional Community is not included in the model because it is collinear with the category dummies. The direct effect of Professional Community is captured by these category dummies, instead. The maximum likelihood estimation (MLE) method is used to estimate the coefficients of the independent variables. We adopt robust standard errors clustered within each contributor to account for possible heteroskedasticity and autocorrelation of errors within contributors in our analyses [17]. It was also important for us to account for differences in the two websites. For example, it may be relatively easier to earn higher points per idea contribution in one community than in the other because of a varying level of voting activities in each community. In our dataset, the average number of points earned per idea is higher in Dell IdeaStorm (mean = 427.2) than in Salesforce.com IdeaExchange (mean = 279.5). Moreover, the average number of comments per user is much greater in Dell IdeaStorm (mean = 118.3) than in 11 Thus, an odds ratio greater than 1 indicates that the organization is more likely to adopt the idea.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

9

Fig. 2. Moderating effect of professional community (idea length and reference pages).

Salesforce.com IdeaExchange (mean = 8.4). Without accounting for such differences, our estimation would have been biased. Any moderating effect found later may reflect a scale effect as well. Therefore, we standardized all the independent variables to zero mean and unit standard deviation within each community for our analysis except for the control variables such as Age of Community and Tenure in Community that do not differ across the two communities and the Category Dummies. 5. Results and discussion 5.1. Estimation results Table 3 shows the estimation results for our study. Column (1) contains the coefficient estimates when only control variables are included. Column (2) adds the main effects of the independent variables. Column (3) shows the estimates with the quadratic effects of supporting evidence added, while Column (4) adds the moderating effect of community type. The pseudo R-squared value, which explains the variance of idea implementation likelihood caused by the antecedents, is 13.23% in Column (2) with the main effects. This increases to 16.08% and 16.68% in Columns (3) and (4), respectively. Referring to Models (3) and (4) with the quadratic and moderating effects, we see that all the hypotheses except H4c are strongly supported. Among the supported hypotheses, the coefficients are significant at 1% level except for H3c,

Table 4 Estimation results of each community separately. Variable

Dell IdeaStorm

Salesforce.com IdeaExchange

Intercept Age of community Tenure in community Same day submission Prior participation Prior implementation rate Idea popularity Idea length Number of reference pages Number of supplementary Images Number of reference pages2 Number of supplementary images2 N Pseudo R-squared

−16.593⁎⁎⁎ (0.232) −0.024⁎⁎⁎ (0.007) 0.012 (0.013) −0.013 (0.083) 0.165⁎⁎⁎ (0.063) 0.062⁎⁎ (0.029) 0.058⁎ 0.031) −0.402⁎⁎⁎ (0.121) 2.715⁎⁎⁎ (0.303) 0.177 (0.124) −1.370⁎⁎⁎ (0.257) −0.025 (0.018) 9984 11.51%

−3.936⁎⁎⁎ (0.592) −0.039⁎⁎⁎ (0.006) 0.009 (0.007) −0.377⁎⁎⁎ (0.090) 0.078 (0.051) 0.139⁎⁎⁎ (0.031) 0.356⁎⁎⁎ (0.071) −0.005 (0.056) 1.701⁎⁎⁎ (0.313) 0.197⁎⁎ (0.094) −0.338⁎⁎⁎ (0.093) −0.027 (0.041) 9980 21.34%

Robust standard errors in parentheses. The coefficients for category variables are not shown for brevity. ⁎⁎⁎ Significant at 1%. ⁎⁎ Significant at 5%. ⁎ Significant at 10%.

which was supported at 5% level in Column (3). We note that the coefficients for the number of reference pages and number of supplementary images increased considerably in Column (3) compared to Column (2) after including their quadratic terms. This increase shows why it is important to include the quadratic terms, as without them, the estimates of these effects may be biased. We mainly refer to Column (4) in Table 3 for subsequent interpretations. We present these results in Fig. 1 as well. It was predicted in H1a and H1b that an innovation idea from a contributor with higher prior participation and prior implementation rate would be more likely to be implemented by the firm, respectively. From the coefficient in Column (4), for every unit increase in the standardized prior participation (number of prior comments made by the idea contributor), the odds that the idea could be implemented increases by 11.3% (β = 0.107, pvalue b 0.05). Next, an increase in prior implementation rate by one standard deviation leads to an increase of odds of implementation by 12.1% (β = 0.114, p-value b 0.01). This result suggests that prior implementation rate of the idea contributor can be viewed by the firm as an indicator of their ability to contribute useful ideas. H2 on the effect of idea popularity was strongly supported (β = 0.235, p-value b 0.01), i.e., an increase in the standardized idea popularity by one standard deviation leads to a 26.5% increase (=(eβ − 1)× 100) in the odds ratio. H3a states that the idea length has a negative effect on the implementation likelihood of the idea. Indeed, an increase in idea length by a unit standard deviation leads to a decrease of odds of implementation by 35.0% (β = −0.431, p-value b 0.01). It was predicted in H3b and H3c that the number of reference pages and number of supplementary images of an innovation idea have an inverted U relationship with its implementation likelihood. These hypotheses are supported by our empirical results in Columns (3) and (4). Using the estimates in Column (3), the idea implementation likelihood increases until the standardized number of reference pages reaches 1.5 and drops off after that. Similarly, the idea implementation likelihood increases until the standardized number of supplementary images reaches 3.212 and drops after that. These results demonstrate that the effects of supporting evidence on idea implementation likelihood are optimal at medium levels of supporting evidence. H4a on the moderating effect of community type was strongly supported in Column (4) (β = 0.423, p-value b 0.01). Fig. 2 shows that the log odds ratio drops more sharply in Salesforce.com IdeaExchange than in Dell IdeaStorm. H4b was also strongly supported (β = 0.703, p-value b 0.01) such that the positive (negative) relationship between the number of reference pages and the likelihood of idea implementation 12 This indicates that idea posting with more than one reference page or two supplementary images in Salesforce.com IdeaExchange would hurt the chances of idea implementation. Similarly, providing more than two reference pages or four supplementary images reduced the chance of idea implementation in Dell IdeaStorm.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

10

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

is greater (lower) in the professional community context than in the hybrid community under study. Fig. 2 also shows that the log odds ratio picks up more sharply in Salesforce.com IdeaExchange than in Dell IdeaStorm when the level of reference page is low to moderate. However, as the level of reference page increases, the log odds ratio drops in both communities while it drops more sharply in Dell IdeaStorm than in Salesforce.com IdeaExchange. However, H4c was not supported i.e., the effect of the number of supplementary images tends to be similar in both communities. Although not hypothesized, it is important to note that the effect of Same Day Submission is negative and significant. From the results, it appears true that the user innovation communities suffer from limited resources and manpower for reviewing such that the likelihood of idea implementation is low when a large number of ideas are submitted at once. Thus, our implicit assumption of the low review capacity in online user innovation communities in this study is credible. The effect of a member's tenure in community is positive and marginally significant. Therefore, a member's learning may accumulate in the community to generate ideas with higher implementation likelihood. Age of community is negatively associated with a firm's likelihood of idea implementation. This indicates that a community may suffer from a decrease in valuable contributions by its members over time, or it may have tightened up its review policy after gaining some experience. As a post-hoc analysis, we tested the model in both communities separately (see Table 4). All the coefficients in both communities are in the same direction. Although the overall pattern of coefficients is similar in both communities, there were a few differences beyond what we hypothesized. First, the effects of most independent variables are more significant in Salesforce.com IdeaExchange than in Dell IdeaStorm. Accordingly, the pseudo R-squared value is larger in the Salesorce.com IdeaExchange than in Dell IdeaStorm. This could indicate that it is more difficult to explain the implementation likelihood in a hybrid community due to unobserved characteristics. Second, the effects of the number of supplementary images in Dell IdeaStorm and prior participation in Salesforce.com IdeaExchange are insignificant. This suggests that there may exist other differences between the two communities beyond those hypothesized.13 Overall, the post-hoc analysis shows the advantage of studying multiple user innovation communities in contrast to prior studies examining one community. The differences between the communities suggest that relying on a single innovation community may suffer from a lack of generalizability. 5.2. Robustness checks To validate our estimation results, we conducted three additional robustness checks with the dataset. First, since the reviewers spend some time to assess an innovation idea, recent ideas might appear less likely to be selected for implementation than older ones. This bias could cause potential inaccuracy in estimation. Considering this factor, we applied the estimation method on the set of ideas contributed at least 6 months before the final data collection. This is because the majority of implementations are decided within 6 months. By excluding the recent ideas, the total number of ideas in our dataset decreased from 19,964 to 16,551.14 However, there was no essential difference between these and the full dataset estimation results. This analysis suggests that our results are robust in this aspect. Second, in our results, it is seen that a longer idea leads to lower idea implementation likelihood. However, it could be argued that the idea length should have a positive effect on the likelihood when the total number of words is below a certain threshold, 13 For example, in line with our reasoning, members in a hybrid community may not always be capable of choosing useful supplementary images. Also, the insignificant result of prior participation in Saleforce.com IdeaExchange may be due to relatively lower participation in Salesforce IdeaExchange (mean = 8.4) than in Dell IdeaStorm (mean = 118.3). 14 In this analysis, we dropped the least contributed category of Salesforce IdeaExchange because the correlation among category variables becomes higher in this reduced sample analysis than in the original dataset.

while it exerts a negative effect when the length is above it. To test for this possibility, we added the square of idea length in the regression, but it was insignificant in the estimation suggesting that our original theorizing was supported. Third, to eliminate possible “fake” submissions, we performed the same analyses with ideas whose length is at least 10, 20, or 30 words. Here, too, we did not encounter any difference in our results. Further, our results might have been biased by some extremely long ideas. To check this, we ran the model without the ideas having more than 200 words, but obtained qualitatively same results. These analyses show that our findings regarding idea length are robust. 6. Implications and contribution Despite an increasing number of user innovation communities being established by firms in recent years, theoretical explanations of a firm's likelihood of implementing a specific user idea have been scant. Motivated thus, this study set out to address three fundamental questions: (1) What user/contributor characteristics influence the implementation likelihood of their ideas by firms? (2) What characteristics of the idea and its presentation influence their implementation likelihood by firms? and (3) How do the effects of idea presentation characteristics differ across the type of user innovation community (i.e., professional vs. hybrid communities)? It provides the following theoretical and practical contributions by addressing these questions. 6.1. Theoretical contributions There are several key theoretical contributions offered by this study. First, with prior research on user innovation communities mostly focusing on exploring users' motivation to contribute [4,23,24,35,37,38], there is limited study of the antecedents of firms' implementation of user ideas. Among the few studies in this area, Di Gangi and Wasko [14] used the diffusion of innovations (DOI) theory to examine how the inherent characteristics of user ideas affect their implementation in Dell IdeaStorm. However, their sample consisted of 21 ideas that are either adopted, or popular and not adopted, unlike our study which uses a large sample of all ideas contributed over a period of time. The other two related studies [4,29] focused on contributors rather than ideas and used learning perspectives to explore how members' idea contribution behavior and their implementation rate change over time. Thus, they examined the characteristics and behavior of the contributor, not the fate of an idea. Our model and dataset not only included idea popularity and contributor characteristics as antecedents but also added its presentation characteristics, e.g. length, references, and images, which we argued and showed are salient in this context. In this manner, we were able to holistically examine the antecedents of idea implementation likelihood in user innovation communities. This helps to answer important questions like does idea popularity matter more than whether the source is an active contributor and commentator in determining its implementation likelihood? If we control for idea popularity and the contributor, do other idea characteristics matter? Second, this study goes beyond DOI theory to add to our understanding of idea implementation in such communities. As per DOI theory, the innovation characteristics, such as relative advantage and compatibility [53], could help to explain the adoption likelihood of the innovation idea. In contrast, this study suggests an alternative view of idea adoption as the persuasion required to implement innovation ideas in a user innovation community. Here, since the idea has not yet been implemented, it may not be easy for firms to assess the innovation characteristics suggested by DOI theory [14]. Further, since the idea is conveyed as a message posting, a reviewer's decision to select the idea can be affected by the message itself, its presentation, and its source or contributor. Thus, given the context of user innovation communities, our study contributes by applying the persuasion perspective to firm's reviewers' innovation idea selection and implementation decision. It finds that firms may look for criteria such as prior participation and prior

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

implementation rate of the contributor in addition to the idea popularity to filter ideas. Third, with the large number of idea submissions to user innovation communities, we posit that the limited resources and manpower [57] as well as reviewers' “bounded rationality” [61] should be taken into account. Indeed, the effects of the idea presentation characteristics in our study reflect how cognitive overload may play an important role in idea selection. As a result, we argue that ideas may not be fully reviewed and thus not be eventually implemented due to the idea's poor message presentation characteristics. For example, one of our key findings based on cognitive overload concepts is that the supporting evidence of the idea in the form of reference pages and supplementary images has an inverted U relationship with its implementation likelihood. Further, the idea length has a negative effect on its implementation likelihood. In sum, our study shows that the alternative perspectives of message persuasion and cognitive overload are valuable in explaining the adoption of user innovation ideas in such communities. Last, it is also notable that prior empirical research on user innovation communities has mainly focused on single websites, e.g., Dell IdeaStorm. This may limit the generalizability of their findings as any heterogeneity across different types of user innovation communities could not been examined. In contrast, our results have provided evidence for the differential impacts of idea presentation characteristics in professional and hybrid communities. We further explored the differences between the two communities in our post-hoc analysis. Overall, the proposed model contributes to our understanding of the phenomenon of innovation idea implementation in user innovation communities with theoretical accounts and sufficient explanatory power using an objectively measured dataset, which addresses various limitations of subjective data. 6.2. Practical implications The establishment of user innovation communities by firms shows that customers can be involved on a large scale in organizational processes such as new product or service development. Our research has several implications for practitioners. First, our study provides guidelines for screening a large volume of contributed ideas for firms starting to implement online user innovation communities. Particularly, firms may not possess sufficient reviewing manpower to process numerous user ideas in a timely manner. Our work suggests that indicators in the form of idea popularity, contributor prior participation, and prior implementation rate can be employed for quick screening of a large number of ideas and then allow for exploratory analysis of potentially valuable ideas that are harder to articulate. Subsequently, the filtered set of ideas can be elaborated on to assess their quality and evidence. Further, idea length and supporting evidence could potentially be turned into heuristics for idea filtering, such that ideas of relatively short length and medium reference pages and supplementary images are retained for further processing. Second, customers/users should be aware that adding supporting evidence to the description of an innovation idea (in the form of reference pages and images) only up to a certain extent increases its likelihood of implementation. Providing support for an innovation idea with a moderate level of such evidence will facilitate understanding of the idea and improve its persuasiveness. Thus, user innovation community practitioners should encourage idea contributors to provide evidence in the form of links and images for their ideas, but not indiscriminately. To achieve this, practitioners can make functions like inserting hyperlinks and images accessible or grant reputation points to the members who include relevant amount of reference pages and images in their contributions. Third, this research finds that the length of the idea has a negative effect on the likelihood of its implementation. Contributors should be aware that the longer the description of an idea is, the less likely it would be selected for implementation, even more so in hybrid communities with more individual customers. This finding

11

along with the result on supporting evidence suggests that relying on text message input may not be the best approach to present and attract innovation ideas from the user innovation community. Current developments in information technologies allow companies to provide richer tools to their community members to develop innovation ideas. It would be a worthwhile investment for practitioners to provide such alternative methods for users to make a contribution. Last, it is important to identify contributors who can potentially submit valuable ideas and respond to their ideas quickly in order to properly incentivize them [29]. The ideas submitted by the users with higher prior participation and prior implementation rate are more likely to be implemented by the firm. This suggests that user innovation communities can benefit from a strategy of retaining their active members and those with high implementation rates by enhancing members' knowledge of products and services of a firm. This knowledge is potentially a precious source of innovation for the firm. Currently, neither Dell IdeaStorm nor Salesforce.com IdeaExchange offers explicit rewards to its customers for contributing useful ideas. Practitioners might profit from devising various ways to better recognize their active users or even offering monetary incentives in order to retain these valuable members in the user innovation community and leverage on their creativity. Most such communities provide a simple leader board by activities or feature a few users. User innovation communities may further benefit by employing other gamification elements (beyond simple points) such as virtual badges and levels to incentivize community members [33]. 6.3. Limitations and future work The study findings should be interpreted in light of its limitations. First, the value of an innovation idea is indirectly measured in terms of the implementation of the idea. The implementation of an innovation idea indicates its perceived potential value, but may not necessarily suggest its commercial value. Further research can be conducted with direct measures of commercial value such as the increased sales volume or increased customer satisfaction by implementing ideas from the user innovation community. Furthermore, although our study focused on the likelihood of idea implementation, another interesting direction for future research would be to study what factors contribute to the time to implement innovation ideas. The duration until implementation can be modeled using a type of survival analysis. In our case, we were not able to model the duration because both communities did not reveal the exact date of idea implementation. Second, it is worthwhile to point out that both Dell and Salesforce.com are from the IT industry. Members in these communities tend to be more IT-savvy and are mostly located in North America. In other industries or cultures, it is possible that users could develop different attitudes towards contributing innovation ideas and present their ideas in a different style. Therefore, the determinants of idea implementation in such communities may be different, which requires further test of the generalizability of our results. Also, our findings may not be generalizable to other online user innovation communities that follow different reviewing procedures. Nevertheless, there are a number of user innovation communities such as MyStarbucksIdea.com that receive many ideas every day and use a similar evaluation process. Our findings and recommendations could be applicable to other communities to the extent that they share common evaluation procedures with the communities in this study. Third, although we controlled for key characteristics of an idea and a contributor, other contributor variables remain unobserved. For example, we may be able to better understand the dynamics of a user innovation community if each contributor's purchase history, demographic data, and attitudes about the brand are provided. Also, there may be some unobservable characteristics of submitted ideas that were not captured in our model. For example, some ideas may not require supporting evidence by their nature. In other cases, contributors may add reference pages or supplementary images that are less relevant to their ideas. It is also possible that some ideas may employ different

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

12

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx

writing styles and include an abstract although they are lengthy. These conditions may matter, but as our study relies on a large scale objective dataset, it is not feasible to assess and code all the idea characteristics. However, we expect that these kinds of differences in ideas are captured by the idiosyncratic error term in our model. Fourth, not all ideas in these communities (especially Dell IdeaStorm) are necessarily about user innovation. Admittedly, there are some users who make irrelevant requests such as asking the firm to lower prices. However, we note that this issue is common in other innovation communities as well. For example, we found more irrelevant requests in other user innovation communities (e.g., MyStarbucksIdea. com) as compared to the ones we studied. Thus, the studied communities are not exceptional in terms of receiving irrelevant idea submissions sporadically. These irrelevant requests do not bias the coefficients as long as they are received in a random manner, i.e., their effects are captured by the idiosyncratic error term in the model. Similarly, we were not able to rule out the possibility of irrelevant votes on submitted ideas. In some cases, an idea may be promoted because it sounds like fun instead of being useful. Last, an issue in applying the idea characteristics identified in this study for screening innovation ideas may be that a few radically innovative ideas may not receive sufficient attention. Future research could be directed towards understanding how to achieve a balance between efficient screening of a large volume of contributed ideas and fully reviewing potentially valuable ideas that are not well understood by other community members. 7. Conclusion Our work sheds light on how to better exploit the potential of user innovation communities. Deriving from theories of message persuasion and cognitive overload, we develop a conceptual model to explain the likelihood of innovation idea implementation based on the characteristics of the submitted idea and its presentation as well as the characteristics of its contributor. Specifically, the contributor's prior participation and prior implementation rate, as well as the idea's popularity, length, and supporting evidence, are found to influence the innovation idea's implementation likelihood. The results also reveal important differences in the effects of idea presentation characteristics for hybrid (i.e., Dell IdeaStorm) versus professional (i.e., Salesforce.com IdeaExchange) user innovation communities. As our study relies on observable characteristics of ideas and members, our findings can be considered in presenting ideas and in designing and improving management of user innovation communities. Overall, this study helps to throw light on the prevalent but little understood phenomenon of user idea implementation in online user innovation communities. Acknowledgements This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF2015S1A5A8016480). This research was partially supported by the Singapore Ministry of Education Academic Research Fund (R-253-000091-112). References [1] V. Andreoli, S. Worchel, Effects of media, communicator, and message position on attitude change, Public Opinion Quarterly 42 (1) (1978) 59–70. [2] Aristotle, The art of rhetoric, Penguin Classics, NY, 1991. [3] A. Baregheh, J. Rowley, S. Sambrook, Towards a multidisciplinary definition of innovation, Management Decision 47 (8) (2009) 1323–1339. [4] B.L. Bayus, Crowdsourcing new product ideas over time: an analysis of the Dell IdeaStorm community, Management Science 59 (1) (2013) 226–244. [5] A. Bhattacherjee, C. Sanford, Influence processes for information technology acceptance: an elaboration likelihood model, MIS Quarterly 30 (4) (2006) 805–825. [6] E. Bridges, R.E. Goldsmith, C.F. Hofacker, Attracting and retaining online buyers: comparing B2B and B2C customers, in: I. Clarke, T.B. Flaherty (Eds.), Advances in Electronic Marketing, Idea Group Inc., Hershey PA, 2005.

[7] R.A. Burgelman, C.M. Christensen, S.C. Wheelwright, Strategic management of technology and innovation, McGraw-Hill/Irwin, Boston, MA, 2008. [8] L. Chen, J. Marsden, Z. Zhang, Theory and analysis of company sponsored value cocreation, Journal of Management Information Systems 29 (2) (2012) 141–172. [9] C.M.Y. Cheung, C.L. Sia, K.K.Y. Kuan, 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) (2012) 618–635. [10] A. Clevinger, Test performance: the influence of cognitive load on reading comprehension, ScholarWorks@Georgia State University, 2014. [11] M. Culnan, P. McHugh, J. Zubillaga, How large U.S. companies can use Twitter and other social media to gain business value, MIS Quarterly Executive 9 (4) (2010) 243–259. [12] R.L. Daft, R.H. Lengel, L.K. Trevino, Message equivocality, media selection, and manager performance: implication for information systems, MIS Quarterly 11 (3) (1987) 355–366. [13] P.M. Di Gangi, M. Wasko, Open Innovation Through Online Communities, in: W.R. King (Ed.), Knowledge Management and Organizational Learning, Springer Publishing Company, London 2009, p. 206. [14] P.M. Di Gangi, M. Wasko, Steal my idea! Organizational adoption of user innovations from a user innovation community: a case study of Dell IdeaStorm, Decision Support System 48 (2009) 303–312. [15] P.M. Di Gangi, M. Wasko, R. Hooker, Getting customers' ideas to work for you: learning from dell how to succeed with online user innovation communities, MIS Quarterly Executive 9 (4) (2010) 213–228. [16] B. Douthwaite, J.D.H. Keatinge, J.R. Park, Why promising technologies fail: the neglected role of user innovation during adoption, Research Policy 30 (2001) 819–836. [17] B. Efron, Nonparametric estimation of standard errors: the jackknife, the bootstrap and other methods, Biometrika 68 (3) (1981) 589–599. [18] B. Faber, Professional identities what is professional about professional communication? Journal of Business and Technical Communication 16 (3) (2002) 306–337. [19] L. Festinger, A Theory of Cognitive Dissonance, Stanford University Press, Stanford, CA, 1957. [20] S. Fournier, S. Lee, Getting brand communities right, Harvard Business Review 68 (4) (2009) 105–111. [21] G.R. Foxall, J.D. Tieney, From CAP1 to CAP2: user-initiated innovation from the user's point of view, Management Decision 22 (5) (1984) 3–15. [22] N. Franke, S. Shah, How communities support innovative activities: an exploration of assistance and security software, Research Policy 32 (2003) 157–178. [23] J. Füller, K. Matzler, M. Hoppe, Brand community members as source of innovation, Journal of Product Innovation Management 25 (6) (2008) 608–619. [24] J. Füller, H. Mühlbacher, K. Matzler, G. Jawecki, Consumer empowerment through internet-based co-creation, Journal of Management Information Systems 26 (3) (2010) 71–102. [25] T. Graham, T. Witschge, In search of online deliberation: towards a new method for examining the quality of online discussions, Communications 28 (2) (2003) 173–204. [26] J. Habermas, The Structural Transformation of the Public Sphere, MIT Press, Cambridge, MA, 1989. [27] M.A. Hamilton, P.J. Mineo, A framework for understanding equivocation, Journal of Language and Social Psychology 17 (1998) 3–35. [28] S.R. Hiltz, M. Turoff, Structuring computer-mediated communication systems to avoid information overload, Communications of the ACM 28 (7) (1985) 680–689. [29] Y. Huang, P.V. Singh, K. Srinivasan, Crowdsourcing New Product Ideas under Consumer Learning, Management Science, Forthcoming, 2014. [30] B.A. Huhmann, D.L. Mothersbaugh, G.R. Franke, Rhetorical figures in headings and their effect on text processing: the moderating role of information relevance and text length, IEEE Transactions on Professional Communication 45 (2002) 157–169. [31] B.T. Johnson, A.H. Eagly, Effects of involvement on persuasion: a meta-analysis, Psychological Bulletin 106 (2) (1989) 290–314. [32] Q. Jones, G. Ravid, S. Rafaeli, Information overload and the message dynamics of online interaction spaces: a theoretical model and empirical exploration, Information Systems Research 15 (2) (2004) 194–210. [33] A. Kankanhalli, M. Taher, H. Cavusoglu, S.H. Kim, Gamification: A New Paradigm for Online User Engagement, in: International Conference on Information Systems, (Orlando, 2012. [34] D. Kirsh, A few thoughts on cognitive overload, Intellectica 30 (2000) 19–51. [35] M. Kosonen, C. Gan, H. Olander, K. Blomqvist, My idea is our idea! Supporting userdriven innovation activities in crowdsourcing communities, International Journal of Innovation Management 17 (3) (2013) 1–18. [36] K.R. Lakhani, E. von Hippel, How open source software works: “free” user-to-user assistance, Research Policy 32 (2003) 923–943. [37] J.M. Leimeister, M. Huber, U. Bretschneider, H. Krcmar, Leveraging crowdsourcing: activation-supporting components in IT-based idea competitions, Journal of Management Information Systems 26 (1) (2009) 197–224. [38] M. Li, S.H. Kim, An Empirical Study of Customer Contribution in Online Brand Communities for Innovation, in: International Conference of Information Systems, Saint Louis, MO, 2010. [39] G.L. Lilien, P.D. Morrison, K. Searls, M. Sonnack, E. von Hippel, Performance assessment of the lead user idea-generation process for new product development, Management Science 48 (8) (2002) 1042–1059. [40] D. Livingstone, Adults' informal learning: definition, findings, gaps and future research, +, 21, 2001. [41] T.M. Lowrey, The effects of syntactic complexity on advertising persuasiveness, Journal of Consumer Psychology 7 (2) (1998) 187–206.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004

M. Li et al. / Decision Support Systems xxx (2016) xxx–xxx [42] D. McFadden, Conditional logit analysis of qualitative choice behavior, in: P. Zarembka (Ed.), Frontiers in Econometrics, Academic Press, New York, 1974. [43] D. McFadden, K. Train, W.B. Tye, An application of diagnostic tests for the independence of irrelevant alternatives property of multinomial logit model, Transportation Research Record 637 (1977) 39–45. [44] G.A. Miller, The magical number seven, plus or minus two: some limits on our capacity for processing information, Psychological Review 63 (2) (1956) 81–97. [45] S. Nambisan, P. Nambisan, How To profit from a better “virtual customer enviroment”, MIT Sloan Management Review 49 (3) (2008) 52–63. [46] D.J. O'Keefe, Justification explicitness and persuasive effect: a meta-analytic review of the effects of varying support articulation in persuasive messages, Argumentation and Advocacy 35 (2) (1998) 61–75. [47] D.H. Park, S. Kim, The effects of consumer knowledge on message processing of electronic word-of-mouth via online consumer reviews, Electronic Commerce Research and Applications 7 (4) (2008) 399–410. [48] D.H. Park, J. Lee, I. Han, The effect of on-line consumer reviews on consumer purchasing intention: the moderating role of involvement, International Journal of Electronic Commerce 11 (4) (2007) 125–148. [49] R.E. Petty, J.T. Caciopoo, R. Goldman, Personal involvement as a determinant of argument-based persuasion, Journal of Personality and Social Psychology 41 (5) (1981) 847–855. [50] R.E. Petty, J.T. Cacioppo, Communication and Persuasion: Central and Peripheral Routes to Attitude Change, Springer-Verlag, New York, 1986. [51] R.E. Petty, D.T. Wegener, Attitude change: multiple roles for persuasion variables, in: D. Gilbert, S. Fiske, G. Lindzey (Eds.), The handbook of social psychology, McGrawHill, New York 1998, pp. 323–390. [52] A.G. Robinson, D.M. Schroeder, Ideas Are Free: How the Idea Revolution Is Liberating People and Transforming Organizations, (Berrett-Koehler Publishers, 2009. [53] E.M. Rogers, Difussion of Innovations, Free Press, New York, 2003. [54] E.M. Rogers, R. Agarwala-Rogers, Organizational communication, in: G.L. Hanneman, W.J. McEwen (Eds.), Communication Behaviour, (Addision Wesley, MA, Reading 1975, pp. 218–236. [55] A.F. Rutkowski, C.S. Saunders, Growing pains with information overload, Computer 43 (6) (2010) 94–95 96. [56] H.J. Schau, A.M. Muniz, E.J. Arnould, How brand community practices create value, Journal of Marketing 73 (5) (2009) 30–51. [57] T. Schulze, M. Indulska, D. Geiger, A. Korthaus, Idea assessment in open innovation: a state of practice, ECIS 2012 Proceedings, Paper 149, 2012. [58] S.K. Shah, M. Tripsas, The accidental entrepreneur: the emergent and collective process of user-entrepreneurship, Strategic Entrepreneurship Journal 1 (2007) 123–140. [59] M. Sherif, C.I. Hovland, Social Judgment: Assimilation and Contrast Effects in Communication and Attitude change, Greenwood Press, Westport, CT, 1961. [60] J.A. Short, E. Williams, B. Christie, The social psychology of telecommunications, Wiley International, London, UK, 1976. [61] H.A. Simon, Models of Man, Wiley & Sons, New York, 1957. [62] S. Slaughter, Innovation and learning during innovation implementation: a comparison of user and manufacturer innovations, Research Policy 22 (1993) 81–95.

13

[63] S.W. Sussman, W.S. Siegal, Information influence in organization: an integrated approach to knowledge adoption, Information Systems Research 14 (1) (2003) 47–65. [64] E. von Hippel, Democratizing Innovation, MIT Press, Cambridge, MA, 2005. [65] E. von Hippel, Democratizing innovation: the evolving phenomenon of user innovation, International Journal of Innovation Science 1 (1) (2009) 29–40. [66] E. von Hippel, R. Katz, Shifting innovation to users via toolkits, Management Science 48 (7) (2002) 821–833. [67] S. Whittaker, L. Terveen, W. Hill, L. Cherny, The Dynamics of Mass Interaction, Proceedings Conference on Computer-Supported Cooperative Work, Seattle 1998, pp. 257–264. [68] S.C. Wu, W.C. Fang, The effect of consumer-to-consumer interactions on idea generation in virtual brand community relationships, Technovation 30 (11–12) (2010) 570–581. [69] Y. Yang, P.Y. Chen, R. Banker, Winner Determination of Open Innovation Contests in Online Markets, in: International Conference on Information Systems, Shanghai, Chima, 2011. [70] V. Zwass, Co-creation: toward a taxonomy and an integrated research perspective, International Journal of Electronic Commerce 15 (1) (2010) 11–48. Mingguo Li was a graduate student at Tepper School of Business, Carnegie Mellon University. He received his Master of Engineering degree from Ecole Polytechnique, France, and MS in Information Systems from National University of Singapore. He has been working in financial technology industry after graduation. Atreyi Kankanhalli is an associate professor in the Department of Information Systems at the National University of Singapore (NUS). She obtained her B. Tech. from the Indian Institute of Technology Delhi, M.S. from the Rensselaer Polytechnic Institute, New York, and Ph.D. from NUS. Her research interests are in knowledge management, virtual teams and communities, and IT-enabled innovation in service sectors e.g., e-government and healthcare. Her work has appeared in premium outlets including MIS Quarterly, Information Systems Research, Journal of Management Information Systems, Journal of the American Society for Information Science and Technology, IEEE Trans. on Engineering Management, ACM Transactions on IS, International Journal of Human Computer Studies, and the Proceedings of the International Conference on Information Systems. She serves or has served on several IS conference committees and on the editorial boards of MIS Quarterly, Information Systems Research, IEEE Trans. on Engineering Management, and the Journal of AIS among others. Professor Kankanhalli's work has received a number of awards including the ACM SIGMIS ICIS 2003 Best Doctoral Dissertation Award and the IBM Faculty Award. Seung Hyun Kim is an associate professor of information systems at the School of Business, Yonsei University. He received his Ph.D. and M.S. from Carnegie Mellon University, and his bachelor's degrees from Yonsei University. His primary research interests include economics of information security, knowledge management, and customer relationship management. His work has been published or is forthcoming in leading academic journals including Information Systems Research, MIS Quarterly, and Communications of the ACM.

Please cite this article as: M. Li, et al., Which ideas are more likely to be implemented in online user innovation communities? An empirical analysis, Decision Support Systems (2016), http://dx.doi.org/10.1016/j.dss.2016.01.004