• No results found

The Effect of User Interaction in Online User Innovation Communities: An Evolutionary Perspective

N/A
N/A
Protected

Academic year: 2021

Share "The Effect of User Interaction in Online User Innovation Communities: An Evolutionary Perspective"

Copied!
51
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The Effect of User Interaction in Online User Innovation Communities:

An Evolutionary Perspective

Timko Ogink (2571137)

University of Groningen, Faculty of Economics & Business Duisenberg Building, Nettelbosje 2

9747 AE Groningen, The Netherlands Supervisor: dr. J.Q. Dong Co-assessor: dr. I. Estrada Vaquero

Master Thesis BA SIM 22nd of June, 2015 Word count: 13,352

Abstract

Because individuals are both affected by the social influences of the community, and also shape it by their own behavior, this paper takes an evolutionary perspective on user interaction in online user innovation communities (OUICs). The effect of received comment characteristics on two types of contribution – commenting and ideating – was investigated. Longitudinal data consisting of 161,135 ideas and 318,581 comments from the OUIC of Starbucks (MyStarbucksIdea) indicate that the length of received comments has a negative effect on a users’ ideating behavior and a positive effect on its commenting behavior. Receiving comments from community users with a visible status indicator (high-status) has a positive impact on both a users’ commenting and ideating behavior. No direct relation between comment positivity of received comments and a users’ contribution could be found. However, comment length combined with positivity reverses the negative direct effect on a users’ ideating behavior to a positive effect. Furthermore, the combined effect of received comment length and user status was found to be a significant influencer of a users’ ideating behavior. When a user receives elaborate comments from users without a visible status indicator (no-status) and short comments from high-status users, the direct negative effect of comment length on a users’ ideating behavior is mitigated.

Keywords: user innovation, online communities, crowdsourcing, user feedback, positive

(2)

Introduction

Open innovation, as defined by Chesbrough (2005, p. 2) is ‘the use of purposive inflows and

outflows of knowledge to accelerate internal innovation, and expand the markets for external use of innovation’. Ideas that originally were generated within the boundaries of the company can now originate

from various other settings. Open Innovation evolved to new paradigm for understanding innovation in the organizational science literature, rather than being just another management hype (Chesbrough, 2006). As users are increasingly capable to innovate themselves (Von Hippel, 2005), Online User Innovation Communities (OUICs) become more important for firms in order to capture value from implementing user-generated ideas (Dong & Wu, 2015). OUICs do not only provide firms with the opportunity to absorb specialized knowledge of voluntary customers’ own experiences with current products or services (Füller, Bartl, Ernst & Mühlbacher, 2006), they can also increase customer loyalty (Füller, 2010) and make the collection of market information better, faster, and cheaper than traditional market research (Howe, 2008). Ideas proposed by customers might even outperform those of professional developers (Magnusson, 2009) or company employees (Poetz & Schreier, 2012). Moreover, working with high performing customers, who perform tasks that are regularly done by employees, can be the basis of a sustainable competitive advantage (Tax, Colgate & Bowen, 2006). Commenting and voting on the idea proposals of other users is seen as an important design element of OUICs (Hutter, Hautz, Füller, Mueller & Matzler, 2011). Firms often use these feedback mechanisms to select a limited number of successful ideas for further evaluation, as managers need to concentrate their cognitive capacity to a limited number of issues in order to achieve strategic performance (Ocasio, 1997; 2013).

(3)

interaction with other users originate from the perception of acquired skills and knowledge from others, a sense of belongingness, perceived peer recognition of their expertise and status, or by experiencing joy and fun. However, these positive perspectives underexpose the hazards of OUICs for both firms and users. When users are unsatisfied with the outcomes of participation, perceive unfairness, or have a low sense of community, their participation in the OUIC may cause negative reactions like negative word of mouth to others. It is important to note that a users’ dissatisfaction is often caused by interaction with other users instead of with the firm (Gebauer, Füller & Pezzei, 2013), which makes it particular interesting to study the effect of user interaction.

Increasing the supply of knowledge by stimulating the willingness of users to share their knowledge in the community is one of the biggest challenges for fostering community (Chiu, Hsu & Wang, 2006). This allows for the transfer, accumulation, transformation, and co-creation of knowledge in OUICs (Faraj, Jarvenpaa & Majchrzak, 2011). The main courses of action by which an individual can share knowledge and contribute to the OUIC are, proposing ideas (ideating), and providing feedback on the idea proposals of other users (commenting). Most of the implemented ideas originate from users that proposed multiple ideas over time, however, unfortunately, most of the users in OUICs only post one single idea (Bayus, 2013). This indicates the importance to gain insight in how received comment characteristics impede or increase a users’ ideating behavior. In sum, the positive and negative impact that user interaction can have on both the host firm and OUIC users emphasizes the importance of studying the effect of received comment characteristics on a users’ contribution behavior.

(4)

knowledge freely in the community (Abdul-Rahman & Hailes, 2000). For the purpose of this study OUICs are defined following Dahlander & Wallin (2006) as; ‘distributed groups of individuals focused

on solving a general problem and/or developing a new solution supported by computer mediated communication’ (p. 1246).

Despite the work and effort scholars have invested in studying feedback in OUICs, there is still a gap in the literature that deserves academic attention. User interaction and feedback are investigated in several ways, whereby authors take a literature review approach (Nambisan, 2002), a network approach (Jones & Churchill, 2009), or use survey data to test their hypotheses (Chiu, Hsu & Wang, 2006; Nambisan & Baron, 2007, 2009, 2010; Wu & fang 2010). Both the paper of Chiu, Hsu & Wang (2006) and Wu & Fang (2010) suggested that future research should take a longitudinal approach for analyzing the effect of user interaction on their contribution behavior. To my knowledge no study addressed this gap specifically tailored to the OUIC context. This paper will address the gap by connecting various measures at the comment level to the emerged theoretical field that explains a users’ voluntary contribution behavior in OUICs. More specifically, by providing evidence from the MyStarbucksIdea community, this paper seeks to contribute to the literature in several ways.

(1) Hundreds of papers have been written on how the amount of feedback affects the behavior of an individual (Shute, 2008). Indeed, feedback can change an individuals’ behavior by enhancing learning, also when provided by peers (Liu & Carless, 2006). However, the content of the feedback should be of high quality in order to be an effective influencer of behavior (Shute, 2008). In the OUIC context, more elaborate comments are theorized to be of higher quality (Lampe & Resnick, 2004), but there is no quantitative evidence in the extant literature that that identifies received comment length as an effective influencer of a users’ future contribution behavior.

(5)

behavior (Smither, Barry & Reilly, 1989; Atwater, Roush & Fishthal, 1995; Reilly, Smither & Vasilopoulos, 1996). On the other hand it is also argued that positive feedback is more effective because it enhances self-image, tends to be perceived more accurately, and negative feedback could be rejected or ignored by the receiver (Markus, 1977; Ilgen, Fisher & Taylor, 1979; Greenwald, 1980; Tesser & Campbel, 1983; Hattie, 2013). These contradictions are also found in the online community literature. For example, it is shown that systematic feedback systems in customer support communities retain high quality contributors due to positive feedback, and repel low quality contributors due to negative feedback that they receive from other users (Moon & Sproull, 2008). Contrary to these findings, Bayus (2013) concludes that receiving positive feedback impedes a users’ future contribution based on the finding that the number of future idea proposals of a user declines after the firm announced to implement their idea.

(3) The effect of user status in OUICs has been studied extensively by making use of the social exchange theory. Deriving status benefits from participation is an important reason why users participate and contribute to the community (Wang & Fesenmaier, 2004; Nambisan & Baron, 2009). Another stream of research investigates how reputation affects the behavior of other actors by investigating online buyer seller relations (Resnick & Zeckhauser, 2002; Dellarocas 2005). Despite the effort made in this theoretical field, quantitative evidence of how user status in OUICs affects knowledge sharing behavior of other users by means of interaction remained largely unexplored.

Addressing these issues forms the main contribution of this paper to OUIC literature. Because the theoretical foundations of social psychology are strong and have proven their validity in analyzing individual behavior in online communities (Chi, Hsu & Wang, 2006), a quantitative research approach was chosen. The following research question was used in order to address the outlined contributions;

what are the effects of received comment characteristics on a users’ ideating and commenting behavior in an OUIC? The individual OUIC user was adopted as the unit of analysis in order to answer the question.

(6)

comments from high-status users positively impacts both a users’ commenting and ideating behavior; (3) the negative effect of received comment length on a users’ ideating behavior can be mitigated by short comments from high-status users and elaborate comments from no-status users; (4) positivity of received comments can reverse the negative impact of comment length on a users’ ideating behavior.

This paper mainly draws on the social cognitive theory and its theoretical components; self-efficacy, and outcome expectations. These are used as the theoretical foundation for the hypotheses in the next section. In the third section the methodology of the research is presented and explained, followed by the results in the fourth section. Subsequently, conclusions are drawn and discussed in the fifth section, along with the managerial implications. Finally, the last section elaborates on the limitations of this paper and future research suggestions are provided.

Theory and Hypotheses

(7)

The social cognitive theory argues that humans are both the product as the producers of the social context in which they are embedded (Bandura, 2001). Self-efficacy and outcome expectations play an important role in the social cognitive theory because they determine whether people think self-enhancing or self-hindering in their behavior and thereby to a large extent which actions will be undertaken, how much effort will be invested, and whether failure works motivating or demoralizing (Bandura, 2001). The theory explains that human behavior can be determined by the degree to which an individual thinks he or she is able to do the job and reach a certain level of performance (self-efficacy), and the degree to which an individual expects consequences of these actions (outcome expectations) (Bandura, 1986). In the OUIC context, these are not only related to personal outcomes (personal outcome expectations), but can also be related to a users’ expectations regarding outcomes for other community user (community outcome expectations) (Chiu, Hsu & Wang, 2006). An important notion about self-efficacy and outcome expectations is that they can change over time by influences of the social environment (Bandura, 1977). This process is driven by self-observation, self-judgment, and self-reaction (Bandura, 1986) and makes the theory particularly suitable for analyzing changes in an individuals’ behavior in OUICs.

(8)

tangible recognition as influencers of a users’ sense of community, obligation, efficacy, and self-esteem. To enhance a users’ contribution it is also important that they perceive commonalities with other users such as a shared vision, they can identify with each other, and that they trust each other (Chiu, Hsu, & Wang, 2006). In the same line of reasoning Nambisan & Baron (2007) argue that identification with the community results in enhanced contribution by increasing a users’ perceived social and personal integrative benefits.

Comment Length

(9)

investments because personal outcome expectations of expertise enhancement are a strong incentive for individuals to engage in user interaction and idea contribution (Nambisan & Baron, 2010). Users may expect to derive more learning benefits from receiving high quality feedback on future ideas proposals. Perceived learning benefits may also increase community outcome expectations. Moreover, a user may expect to apply their enhanced skills and knowledge in order to help other community users by commenting on their idea proposals. At the same time, personal outcome expectations may increase due to expected enhancement of expertise or reputation recognition they can derive from their commenting behavior (Chiu, Hsu, & Wang, 2006). In sum, comment length has a higher potential to enhance learning, thereby it enhances a users’ perceived competences, self-efficacy, and outcome expectations and are expected to result in increased contribution to the OUIC in terms of proposing new ideas and commenting on the idea proposals of other community users. Therefore the following if formally hypothesized.

H1: The length of received comments positively influences a users’ future contribution to the OUIC in terms of; (a) the number of idea proposals, and (b) the number of comments.

Commenter Status

(10)

Schaar & Zhang, 2014). The reputation signals of the status indicators in OUICs provide comment recipients with the opportunity to easily judge the credibility of the comment (Brown, Broderick & Lee, 2007) and thereby reduce uncertainty and information asymmetry between users (Ba, 2001). Feedback form a trustworthy source has a higher potential to enhance learning because recipients take it more seriously (Shute, 2008).

Next to deriving learning benefits from received feedback, users can also learn by commenting on the idea proposals of other users. This helps them to develop objectivity related to community standards, and, subsequently, is useful for their own contribution whereby their knowledge is extended to a more public level in the community (Liu & Carless, 2006). In the same line of reasoning, comments from users that already launched an idea may be valued because they have knowledge about expectations of the host firm. In other words, active users know more about the community, the firm, and their expectations, and subsequently transfer their enhanced expertise by commenting and ideating. Accumulated comments, votes, and ranks give comment recipients therefore a hunch about the quality of the commenter (Resnick, Kuwabara, Zeckhauser & Friedman, 2000). Moreover, comments from high-status users in OUIC offer more potential for learning as their quality is higher. Just like with enhanced learning by receiving elaborate comments it is expected to increase a users’ self-efficacy (Nambisan & Baron, 2007), personal outcome expectations, and community outcome expectations (Chiu, Hsu, & Wang, 2006).

(11)

and personal integrative benefits, which has a positive impact on perceived self-efficacy and outcome expectations, and therefore enhances a user’s contribution to the OUIC.

H2: Commenter status of received comments is positively influences a users’ future contribution to the OUIC in terms of; (a) the number of idea proposals, and (b) the number of comments.

Comment Positivity

Next to potential learning benefits users may derive from received comments, they may also derive certification (summative) benefits (Becker & Hughes, 1995). These authors argue that they stem from verification of high levels of performance when an individual receives positive feedback. Thereby it enhances an individuals’ self-image and leads to a better mood (Brown, Collins & Schmidt, 1988). A positive comment indicates that the commenter has a positive opinion on the potential implementation of the idea proposal of the iterator, whereas a negative comment indicates the opposite (Hu & Liu, 2004). Positive feedback is an important influencer of personal integrative benefits such as peer recognition, a sense of credibility, and confidence because a user wants to be seen as skilled and knowledgeable by other community users (Jeppesen & Molin, 2003; Nabisan & Baron, 2007; 2009). Because peer recognition is also related to experiencing joy (Jeppesen & Molin, 2003) and emotional wellbeing (Ryan & Deci, 2001) it is found to increase a users’ perceived hedonic benefits (Nambisan & Baron 2007). Moreover, by experiencing these positive goal outcomes of participation, self-efficacy, self-regulation and goal setting positively reinforce each other over time (Schunk, 1990). Personal and community outcome expectations of future effort investment are likely to increase because users may expect to receive more positive comments on future contributions and help other community users with their relevant knowledge (Chiu, Hsu & Wang, 2006).

(12)

2006), and identification with the community of the comment recipient (Nambisan & Baron, 2007; 2010). Thereby a users’ community outcome expectancies of helping others’ in de community also decreases along with its self-efficacy and personal outcome expectations. In sum, because receiving positive feedback enhances a users’ perceived personal integrative and hedonic benefits, and subsequently self-efficacy and personal outcome expectations, it is expected to increase their future contribution to the OUIC. Additionally, the negative effect of negative feedback on a users’ perceived shared vision and community identification is likely to result in a decrease of a users’ self-efficacy and outcome expectations. Therefore the following is hypothesized.

H3; Positivity of received comments positively influences a users’ future contribution to the OUIC in terms of; (a) the number of idea proposals, and (b) the number of comments.

Comment Length and Commenter Status

(13)

expected that high-status users do not need to elaborate as much compared to no-status users. This implies that there is less need for high-status commenters to demonstrate their expertise by posting elaborate comments and allows them so stick to the core message. Moreover, the combination of trust and message quality is expected to result in more efficient learning of the comment recipient. The opposite is expected for received comments from no-status community users. These users may post elaborate high quality comments to overcome their lack of trust and quality signals compared to high-status users. Thus in sum, it is expected that a users’ perceived learning, social integrative, and personal integrative benefits from participation in the OUIC increase even more by receiving non-elaborate comments from high-status users and elaborate comments from no-high-status users.

H4: Commenter status of received comments negatively moderates the relation between the length of received comments and a users’ contribution to the OUIC in terms of; (a) the number of idea proposals, and (b) the number of comments.

Comment Length and Positivity

(14)

the direct effect of comment length. Thereby the joint effect has a larger impact on a users’ outcome expectations and self-efficacy. Additionally, positivity and comment length combined can have an effect on multiple perceived benefits at the same time. Next to enhancing a users’ perceived learning benefits, they can also increase its perceived hedonic and personal integrative benefits due to the direct effect of comment positivity (Jeppesen & Molin, 2003; Nabisan & Baron, 2007; 2009). In sum, elaborate comments combined with positivity are expected to enhance a user’s contribution behavior because they are likely to result in higher learning benefits compared to the direct effect of comment length, and because they address multiple motivational factors that drive self-efficacy and outcome expectations at the same time. Therefore it is hypothesized that positivity of received comments positively moderates the relation between received comment length and a users’ future contribution to the OUIC.

H5: Positivity of received comments positively moderates the relation between the length of received comments and a users’ contribution to the OUIC in terms of; (a) the number of idea proposals, and (b) the number of comments.

(15)

Figure 1. Conceptual model

Methodology

Research Setting

(16)

were submitted offline and thus not posted by a real OUIC user. The final dataset contained 161,135 ideas and 318,581 comments posted between 26-03-2008 and 23-02-2015.

Data

(17)

unique user-days. Because only the days in which a user received feedback had to be analyzed and tested on a users’ contribution behavior in the next time window, the total number of observations could be reduced to 88,026. Before deleting these observations, the contributing behavior of ideating and commenting was derided from the next user-day.

Measures

In order to distinguish between variables that measure received and contributed interaction, the indicators r_ (for received) and c_ (for contributed) were added to the variable names. In example the average number of received words per comment was denoted as r_words and the number of provided comments on the idea proposals of other users as c_comments.

Dependent variables

To measure a users’ future contribution in terms of commenting and ideating in the next time window, both count-based measures were adopted from Bayus (2013). For ideating the measure c_ideas was created to measure the total number of idea proposals in the next time window for each user. To measure future commenting activity, c_comments was introduced as a measure of the total number of

provided comments on the idea proposals of other users.

Explanatory variables

(18)

To measure comment positivity, the variable r_positive was introduced. In order to generate a score for each comment the count based text-mining method of Meyer, Hornik and Feinerer (2008) was adopted. To implement their method a PHP: Hypertext Preprocessor (PHP) script was created. Before the text mining analysis was conducted all punctuation marks had to be removed from the comments in order to isolate full words between spaces. This script selected all words separately in a Structured Query Language (SQL) column and row. Subsequently it counted how many words appeared in either a list of positive or negative words, resulting in two count based values per comment; the number of positive words (P) and the number of negative words (N). These were converted to a positive-ratio per comment; 𝑃 (𝑃 + 𝑁). Subsequently the average of these ratios were taken over the total number of received comments per user-day in data aggregation. In sum, the measure represents the average positivity ratio of received comments per unique user-day. The lists of positive and negative words for the text mining analysis were adopted from (Hu & Liu, 2004). These lists were created specifically for text mining analysis of customer comments in order to gauge whether they are positive or negative about current firm offerings. This makes the analysis particularly suited to measure whether the commenter has a positive or negative opinion on the idea proposals of other users in the community.

(19)

Interaction variables

Two interaction variables had to be calculated based on the explanatory variables. For the joint effect of commenter status and comment length, r_status and r_words were multiplied with each other in order to create r_statusXr_words. To calculate the variable for the joint effect of comment positivity (r_positive) and comment length (r_words), also these variables were multiplied in order to create

r_positiveXr_words.

Control variables

Timing is seen as an important variable in the feedback literature (Hattie & Timperly, 2007), and also in online communities a quick response from other users is important (Wooten & Ulrich, 2011). Receiving a quick response from other community users has a positive impact on both a users’ perceived hedonic and social integrative benefits (Nambisan & Baron, 2009). Therefore the number of days between the idea proposal and the comment was measured by the variable r_age and integrated as a control variable in this study. In aggregation the average was taken over the total number of received comments per user-day. Thereby the variable r_age represents the average age of received comments per unique user-day.

(20)

Statistical analysis

In order to generate descriptive statistics, several tests were conducted in Stata 13.1. First, the distributions of all variables were inspected after generating a histogram including a normality curve. Second, the summary statistics were generated for all variables. Third, the skewness kurtosis test for normality was applied to control for a normal distribution. These tests indicated that the distribution of the control variable r_comments and the explanatory variable r_words could be improved by logarithmic transformation, which was applied before conducting the main analysis. The paragraph that presents the descriptive statistics in the next section presents data regarding the original variables. The indicator ln was added to the variables in models for which the logarithmic transformation was used. Fourth, Spearman’s rank correlation test was conducted to measure correlations among variables and along with the test for variance inflation factors (VIFs) was used to check for multicollinearity. Finally, some calculations were made based on the non-aggregated data to present descriptive statistics on MyStarbucksIdea. Both the skewness kurtosis test for normality and the VIF test were based on based on the residuals of a multiple OLS (ordinary least square) regression where all explanatory and control variables were included in the model and tested on both dependent variables.

Multiple OLS regression was not used for the main analysis of this paper because the dependent variables are count based, therefore not continuous, and also do not follow a normal distribution. Thereby some main assumptions of multiple OLS regression were violated, and two alternatives for count based dependent variables better suited; Poisson regression and negative binominal regression (Hilbe, 2011). In order to make a choice between the two, the summary statistics in table 1 were analyzed.

(21)

unobserved user specific characteristics that correlate with one or more of the independent variables. The dataset can be titled as unbalanced due to differences in the number of observations per user. The fixed effects model however cannot make calculations for users that were only observed once, and these observations would have been dropped automatically during statistical analysis. Because dropping these observations would create a large sampling bias, the random effects model was used for statistical analysis, and the assumption that there are no unobserved user specific characteristics that correlate with one or more of the independent variables was accepted. Subsequently, the same test was conducted to measure the standardized correlation coefficients of the model in order to compare the impact of different received comment characteristics.

Results

Descriptive Statistics

The summary statistics of all included variables are presented in table 1. The skewness kurtosis test for Normality indicates that none of the residuals from the two models follow a normal distribution (p < 0.01). By observing the histograms of the distributions could also be verified that the residuals of the models with respect to both dependent variables are right skewed.

Table 1. Summary statistics (user-day level)

Observations Mean Std. dev. Min Max

(22)

Spearman’s rank correlation test was used to predict correlations among all variables that were used for statistical analysis (Zar, 1972). To prevent a type 1 error, these results were studied along with the VIFs in order to check for multicollinearity problems. The results of the Spearman’s rank correlation test are presented in table 2. The results of the VIF analysis indicate an average VIF of 1.10 with a maximum of 1.12 for the tests on both dependent variables. Spearman’s rank correlations test and the VIFs indicate no multicollinearity problems. For the VIFs, as a rule of thumb, a score above 10 is seen as problematic (O’brien, 2007).

Table 2. Spearman’s rank correlation test

(1) (2) (3) (4) (5) (6) (7) (1) R_age 1 (2) R_comments -0.1973* 1 (3) R_status -0.4249* 0.3751* 1 (4) R_words 0.0554* 0.0895* -0.0579* 1 (5) R_positive 0.0846* -0.0678* -0.1192* 0.2608* 1 (6) C_ideas -0.0676* -0.0016 0.0403* -0.0180* -0.0088* 1 (7) C_comments -0.0475* 0.0218* 0.0585* 0.0025 -0.0043 -0.0358* 1 *Significant (p < 0.05)

(23)

Figure 2. Proposed ideas in MyStarbucksIdea Figure 3. Posted comments in MyStarbucksIdea

The comments in MyStarbucksIdea are more positive than negative with an average positivity ratio of 0.66. The average comment length is approximately 44 words. No-status users posted most of the ideas in MyStarbucksIdea; high status users posted 10,225 ideas compared to 115,883 ideas of no-status users. With respect to commenting, the contribution to the community is almost equally divided between the two status groups. High-status users provided approximately 49% of all the comments.

Several t-tests were conducted to check for any significant differences in the behavior of both status groups by comparing the means of comment positivity and comment length. The results show that high-status users were significantly less positive with a mean of 0.64 compared to a mean of 0.70 for no-status users. Furthermore, they used significantly fewer words in their comments, on average 41.35 compared to 47.15 for no-status users. All t-test statistics are significant at p < 0.01.

Hypotheses Testing

The regression results of the negative binominal regression test for panel data are presented in Table 3. Model 1 presents the direct effects whereas in model 2 the hypothesized moderators were added.

A positive effect of comment length on a users’ contribution behavior was predicted by the first hypothesis. Contrary to the expectations, the effect of comment length on a users’ ideating behavior

(24)

shows a significant (p < 0.05) negative correlation coefficient (-0.038). Comment length however does have a significant (p < 0.1) positive effect on a users’ commenting behavior. Thus, the results indicate a significant opposing effect of comment length on the two types of a users’ contribution behavior. Both effects remain significant in model 2. Therefore hypothesis 1a was not supported whereas hypothesis 1b was supported.

The results of the second hypothesis verify that commenter status is positively related to a users’

ideating and commenting behavior, by showing a significant (p < 0.01) positive effect on both dependent

variables. All effects remain significantly positive in model 2. Based on these results, hypothesis 2a and 2b were supported.

The third hypothesis predicted a positive effect of comment positivity on both types of a users’ contribution behavior. However, none of the results could confirm any of these expectations. Comment

positivity is not significantly (p > 0.1) related to a users’ ideating and commenting behavior, therefore

hypothesis 3 was completely rejected.

Hypothesis 4 predicted that commenter status negatively moderates the relation between

comment length and a users’ future contribution behavior. The results show a negative correlation

coefficient (-0.934) and significant impact (p < 0.05) on a users’ ideating behavior. The direct effect of

comment length on ideating has a regression coefficient of -0.375, and of -0.041 in model 2. Although a

positive direct effect was hypothesized, the interaction variable still has a more negative effect compared to the direct effect, in line with hypothesis 4a. The same interaction effect is not significantly related to a users’ commenting behavior, indicating that hypothesis 4b was not supported.

The theoretical framework of hypothesis 5 theorized a positive interaction effect of comment

positivity and comment length on future contribution of the comment recipient. The results indicate that

the effect of comment length on ideating is also significant (p < 0.05) and positively moderated by

(25)

results are in line with hypothesis 5a. However, the interaction effect could not be confirmed to have an effect on a users’ commenting behavior, and therefore hypothesis 5b was rejected. In sum, both moderators have a significant effect on a users’ ideating behavior, but not on commenting behavior. In order to find stronger support regarding hypothesis 4a and hypothesis 5a the standardized correlation coefficients were analyzed to compare their impact with the direct effect of comment length.

Table 3. Results of the negative binominal regression for panel data

Model 1 Model 2

C_ideas C_comments C_ideas C_comments

R_age -.0010072*** -.0012336*** -.0010141*** -.0012384*** R_comments (ln) -.1961318*** -.1206347*** -.1827789*** -.1179123*** R_status .2008189*** .2732388*** .5034054*** .4966935*** R_words (ln) -.0375239** .0408068* -.0505718* .0813459** R_positive .0180925 .0721186 -.3035934** .1837887 R_statusXlnwords -.0934299** -.0671684 R_positiveXlnwords .1026375** -.035264 Cons .0487137 -1.184164*** .0695133 -1.314319***

***Significant at p < 0.01; **significant at p < 0.05; *significant at p < 0.1

Standardized test

The results of the standardized correlation coefficients (presented in table 4) indicate that the control variable regarding the age of received comments has the largest impact on both a users’

commenting and ideating behavior. From the explanatory variables, the effect of commenter status of

received comments has the largest impact on a users’ contribution behavior compared to all others. The standardized coefficients of the moderators show that length combined with positivity of received comments is a stronger influencer of a users’ contribution compared to its combination with commenter

(26)

effect of comment length on a users’ ideating behavior. These findings provide strong support for hypothesis 4a and hypothesis 5a.

Table 4. Standardized results of the negative binominal regression for panel data

Model 1 Model 2

C_ideas C_comments C_ideas C_comments

R_age -.3478142*** -.4259935*** -.3501917*** -.4276409*** R_comments (ln) -.1038111*** -.0638511*** -.0967435*** -.0624102*** R_status .0696295*** .0947395*** .1745447*** .1722174*** R_words (ln) -.0373217** .0405869* -.0502992* .0809075** R_positive .0071561 .0285249 -.1200796** .0726935 R_statusXlnwords -.1140836** -.0820167 R_positiveXlnwords .1535265** -.0527484 Cons -.237682*** -1.170687*** -.2330844*** -1.171093***

***Significant at p < 0.01; **significant at p < 0.05; *significant at p < 0.1

Controls

The control variable age of received comments has a significant negative effect (p < 0.01) on both a users’ ideating and commenting behavior. The results remain significantly negative in model 2, indicating that a quick response from the community has a positive effect on future both types of a users’ future contribution behavior. These findings are in line with the extant literature (Nambisan & Baron, 2009; Wooten & Ulrich, 2011).

The second control variable, number of received comments, has a significant negative effect on

ideating and commenting behavior (p < 0.01), which is also confirmed by Model 2. Thereby, these

(27)

interactions is positively related to idea generation (Wu & Fang, 2010), and social interaction ties are positively related to the quantity of knowledge sharing (Chiu, Hsu & Wang, 2006). However, by analyzing the surveys of these authors in more depth it becomes clear that they not only measured the sheer quantity of interaction. Nambisan & Baron (2009) include time and firm interaction in the measures of their construct, Wu & Fang (2010) integrate – next to frequency of interaction – the scope of interaction and the mode of interaction in their measures, and Chiu, Hsu & Wang (2006) place importance on the strength of social relations by measuring tie strength. Based on these comparisons, a possible explanation for the negative relation is that sheer quantity of interaction only is not enough to enhance a users’ contribution and may even have a negative impact. Some support for this line of reasoning could be found. Wooten & Ulrich (2011) state that most of the feedback in OUICs is often unrelated to the actual demands of the host firm, conceptualized as ‘random feedback’. They argue that the effect of random feedback is even worse than receiving no feedback at all, and state that it makes an individual perceive their efforts as being wasted. Moreover, deviating interests between the community and the focal user will result in a lower identification with the community (Nambisan & Baron, 2007; 2010) and a lower perception of a shared vision (Chiu, Hsu & Wang, 2006).

Robustness Checks

To verify whether the results are robust, the same test was performed with data aggregated on the user-week level and two sub samples of high-status and no-status users were tested. The first test verifies whether the hypothesis are also supported with less sensitive data whereas the second test checks whether received feedback has a different impact on high-status and no-status users.

Aggregation per week

(28)

significant opposing effect (negative) was found regarding the effect of commenter status on a users’ commenting behavior. Nambisan & Baron (2009) argue that such a difference can be explained because a larger time lag exists between perceived learning benefits and a users’ ideating behavior compared to perceived learning benefits and a users’ commenting behavior. Therefore it may be that the benefits of received comments from high-status users only have a short impact on a users’ commenting behavior and the larger time lag could sustain the positive effect on a users’ ideating behavior. Moreover, the robustness check regarding the weekly aggregated data verifies that the size of the time window used for data aggregation is important, as less sensitive data may reveal less relations, or even expose a different impact of variables. In example, both moderating relations could not be supported in the weekly aggregated model.

Table 5. Results of the negative binominal regression for panel data (user-week aggregation)

Model 1 Model 2

C_ideas C_comments C_ideas C_comments

R_age .0006859*** -.0003629*** .0006857*** -.0003633*** R_comments (ln) -.4534972*** .3715178*** -.4518284*** .3721161*** R_status .2141145*** -.4875925*** .2298144** -.4206434*** R_words (ln) -.0483578*** .0193556** -.0568444*** .0226521 R_positive -.0402268 .0374594 -.1079336 .043639 R_statusXlnwords -.0049385 -.0195254 R_positiveXlnwords .021651 -.0019409 Cons .0953973 .1368882*** .1183785* .1258068**

***Significant at p < 0.01; **significant at p < 0.05; *significant at p < 0.1

Subsample

(29)

distinguish the different effects of the comment characteristics between the two groups. The results for no-status users are presented in Table 6 and the results for high-status users in Table 7. The robustness check revealed some interesting findings.

Interaction characteristics have a much larger impact on future contribution of no-status users compared to high-status users. High-status users may have higher levels of pre-existing self-efficacy, which may be an explanation of why the effect of received comment characteristics is less pronounced (Nambisan & Baron, 2009; Hattie, 2013). Moreover, due to the higher self-efficacy they developed over time, they may not be impacted by short-term commenting behavior of other community users. High-status community users may also be more motivated to sustain the personal integrative benefits they derive from their high status, despite the feedback they receive from others (Nambisan & Baron, 2009).

Comment length only has a negative effect on high-status users’ ideating behavior, and only a

positive effect on no-status users’ commenting behavior. The robustness check revealed a possible explanation of why a counterintuitive relation was found regarding the impact of comment length on a users’ ideating behavior (hypothesis 1a) in the main analysis. It could be that although elaborate comments are associated with higher quality (Lampe & Resnick, 2004; Blumenstock, 2008), their quality is still not high enough to increase perceived learning benefits of high-status users. Moreover, when users receive comments that are perceived by them as being useless, they can get the impression that they wasted their effort of idea generation (Wooten & Ulrich, 2011). This argument could also be a good explanation of why the moderators could not be supported for high-status users, but only are applicable to no-status users.

(30)

status indicators in one measure. In example, top commenters may still learn from users with the idea launched status.

Table 6. Results of the negative binominal regression for panel data no-status subsample

Model 1 Model 2

C_ideas C_comments C_ideas C_comments

R_age -.0010631*** -.0011982*** -.001072*** -.0012002*** R_comments (ln) -.2394228*** -.095855* -.2242937*** -.0935007* R_status .2214569*** .3445279*** .5932254*** .5374176** R_words (ln) -.027692 .0671581** -.0325668 .1010549** R_positive .0373425 .0705884 -.254129 .178271 R_statusXr_wordsln -.1151483** -.0577328 R_positiveXr_wordsln .0931817* -.0336716 Cons -.0499333 -1.717924*** -.0522892 -1.8269***

***Significant at p < 0.01; **significant at p < 0.05; *significant at p < 0.1

Table 7. Results of the negative binominal regression for panel data high-status subsample

Model 1 Model 2

C_ideas C_comments C_ideas C_comments

R_age -.0009874*** -.0015107*** -.0009807*** -.0015197*** R_comments (ln) -.1151103* -.1793924*** -.1071605* -.1797497*** R_status .1841511* .1296157 .1924934 .2953856 R_words (ln) -.0805362** .0041874 -.1296966* .0521527 R_positive -.0868431 .0815621 -.4895448 .2736024 R_statusXr_wordsln -.00135 -.0505068 R_positiveXr_wordsln .1284642 -.0611028 Cons .3226131* -.6791196*** .448463* -.8245775***

(31)

Table 8 provides an overview of all hypotheses of this paper, indicates the nature of the relation, and presents whether a hypothesis is supported or not. From in total ten hypotheses, five could be supported, the other half was rejected, from which one indicated a significant opposing relation.

Table 8. Overview of hypotheses

Hypothesis Independent variable Relation Dependent variable Result

H1a Comment length + Number of idea proposals Not supported*

H1b Comment length + Number of comments Supported

H2a Commenter status + Number of idea proposals Supported

H2b Commenter status + Number of comments Supported

H3a Comment positivity + Number of idea proposals Not supported

H3b Comment positivity + Number of comments Not supported

H4a Status X length - Number of idea proposals Supported

H4b Status X length - Number of comments Not supported

H5a Positivity X length + Number of idea proposals Supported

H5b Positivity X length + Number of comments Not supported

*Significant opposing relation found

Discussion and Conclusion

(32)

First, just the velocity of information in received comments, measured by the length of comments, is not enough to enhance a users’ ideating behavior. Instead, contrary to the expectations, a negative effect was found. It may be that despite longer comments are more informative, the receiver may not be willing to invest the effort to read and understand it (Hsu, Khabiri & Caverlee, 2009; Khabiri, Hsu & Caverlee, 2009). Another explanation is provided by the robustness check regarding the sub-sample. The results indicated that the negative effect was only significant for high-status community users. These results suggest that the average quality of a high-status community user is higher than the average quality of elaborate comments. Thereby these comments can make a high-status contributor perceive their ideating efforts as being wasted (Wooten & Ulrich, 2011), have a negative effect on a users’ self-efficacy and outcome expectations, and decreases its ideating behavior. The expected positive effect on a users’ commenting behavior however was supported, indicating that users perceive a difference between contribution to the community (commenting) and contribution to the company (ideating). A possible explanation of the different effects is that a larger time lag between expertise enhancement and contribution to the firm in terms of product development may exist, compared to contributions to the community in terms of commenting (Nambisan & Baron, 2009).

Second, when received comments are provided by high-status users both ideating and commenting behavior of the recipient increase due to the perception of social integrative benefits and perceived learning benefits due to the higher comment quality, which is in line with the theoretical framework.

(33)

play a role in processing negative or positive feedback (Nambisan & Baron, 2009). Users with low pre-existing self-efficacy may interpret positive feedback still as inefficiencies that need to be overcome and people with high self-efficacy may make more optimistic predictions after receiving negative feedback (Bandura, 2001; Hattie, 2013).

Elaborate comments increase knowledge sharing in the community by stimulating a users’ commenting behavior, but fail to translate the increased knowledge sharing into idea generation. According to the social learning theory (Bandura, 1977), receiving elaborate comments results in providing elaborate comments because individuals copy the interaction behavior of others, resulting in a potential vicious circle of increased knowledge sharing and decreased ideating. Breaking the chain of decreased ideating without negatively impacting commenting behavior of users can be a fruitful way for firms to increase the performance of their community. There are two findings that can play an important role in this case and represent the third and fourth main contribution of this paper.

Third, the first finding indicates that comment positivity can mitigate the negative effect of elaborate comments on a users’ ideating behavior by turning it into a positive effect, without influencing its commenting behavior. In other words, if received comments are both elaborate and positive, users increase their ideating behavior, which is in line with the theoretical framework in which was argued that the combination of perceived benefits jointly increase self-efficacy and outcome expectations.

(34)

from received comment characteristics have a different impact on a users’ community outcome expectations compared to personal outcome expectations and thereby translate to differences in a users’ commenting and ideating behavior. In sum, the findings of this paper contribute to the literature by showing how received comment characteristics impact a users’ commenting and ideating behavior in OUICs.

Managerial implications

Because some managerial implications require a trade-off, this section aims to highlight these trade-offs and provide managers with valuable input for their decision-making process.

Managers should aim to increase – and specifically their commenting behavior – the contribution of high-status community users, because receiving comments from these users has a strong positive impact on a users’ contribution behavior. The downside however is that it also increases the commenting behavior of no-status community users, which subsequently can have a negative impact on the contribution behavior of other users.

Managers may also consider integrating external status indicators into the community. For example status indicators for highly educated individuals or specialized employees. Due to the personal integrative benefits they can derive from these status indicators it may also be a motivation for these potentially highly valuable individuals to join the community (Nambisan & Baron, 2009). The practical implementation leaves room for managerial creativity, but one could think of status indicators like ‘marketing specialist’, ‘financial specialist’, ‘innovation specialist’ etc. However, a good screening protocol should be in place because without certification system these status indicators work inefficiently (Van der Schaar & Zhang, 2014).

(35)

increased without influencing their commenting behavior. Therefore the implications regarding positivism are quite straight forward and do not trigger trade-offs.

A word limit for comments can reduce the negative direct effect of comment length on the ideating behavior of high-status users and reduce commenting behavior of no-status users. However, there are two trade-offs regarding the word limit. The first is that no-status users cannot compensate their lack of credibility and trust in the community by writing elaborate comments. This issue could be addressed by implementing the word limit for high-status user only. The second is that the positive effect on a users’ ideating behavior by receiving positive and elaborate comments cannot be sustained. Because the direct effect of comment length on ideating behavior is much weaker than the interaction effects, one might also argue that a word limit is a bad idea. Perhaps the safest way is to stimulate no-status users to post elaborate comments.

(36)

Limitations and Future Research

In this section the main limitations of this paper are discussed and suggestions for future research are provided. Starting with the first and perhaps most important limitation, the generalizability of the results. Despite the fact that this paper does not draw on sample data, but uses all interaction data over the complete lifetime of MyStarbucksIdea, generalizability to other community contexts may be questionable. Interaction is rooted in the current offerings of the firm and also the nature of the technology that shapes the OUIC environment plays role (Nambisan & Baron, 2009). Future research can be aimed at generalizing the findings of this paper to various OUIC contexts by investigating the OUICs of firms in various industries.

The second limitation is that the status variable could not be derived from the MyStarbucksIdea community with the same time specific accuracy compared to the other variables. Moreover, the status indicators ‘top commenter’ and ‘idea launched’ in MyStarbucksIdea were also added to the interactions before a user earned the status indicator. However, it is not likely that this limitation influences accepting any status related hypothesis in a positive manner, it is more likely to find less expressed results. Despite this limitation, most hypotheses regarding the effect of user status were supported. There is only a chance that the hypothesis based on interaction with the comment length and its effect on a users’ commenting behavior fell prey to a type 1 error and got falsely rejected, and that conclusions or implications regarding the impact magnitude of the status variable are understated. Future research can address this limitation by analyzing different communities on which more accurate time specific status variables are available.

(37)

It is likely that a minor amount of idea proposals and their comments could not be derived from the website of MyStarbucksIdea, which forms the fourth limitation. In popular categories, the maximum number of pages was reached and no further ideas were presented. The limitation was addressed by scraping the website in multiple ways. More specifically, by changing the order in which ideas were presented in four different ways and redo the scraping process, deleting the duplicates afterwards. Eleven out of fifteen categories could be scraped at once because the maximum number of presenting pages was not reached. Four categories had to be scraped in multiple ways, from which for one category it is likely that not all ideas and comments could be collected.

Furthermore, the robustness checks and control variables indicate some fruitful directions for future research. The robustness check identified clear differences in the effect of comment characteristics on the different status groups. Possible explanations were provided in the discussion, however, more research is needed to be conclusive about this phenomenon. Also the counter intuitive results regarding the negative impact of the number of received comments (control variable) on both a users’ ideating and commenting behavior indicate that more research is needed to draw conclusions from these findings.

Questions remain about the duration of the various effects of comment characteristics. As the robustness checks indicate, some differences between the weekly aggregated data and daily aggregated data were found. Future research could look at this issue in more detail.

(38)

Acknowledgements

I learned a lot during the process of writing this master thesis and look back at an enjoyable period, which, I attribute in a large part to the cooperation with my supervisor Dr. John Qi Dong and my fellow students Esteban Barnhardt, Marc Donders and Jork Netten. Their valuable comments and helping hand have contributed a lot during this process.

References

Abdul-Rahman, A., & Hailes, S. (2000, January). Supporting trust in virtual communities. System

Sciences, 2000. Proceedings of the 33rd Annual Hawaii International Conference on (pp. 9-pp). IEEE.

Atwater, L., Roush, P., & Fischthal, A. (1995). The influence of upward feedback on self-and follower ratings of leadership. Personnel Psychology, 48(1), 35-59.

Ba, S. (2001). Establishing online trust through a community responsibility system. Decision Support

Systems, 31(3), 323-336.

Bandura, A. (1977). Self-efficacy: toward a unifying theory of behavioral change. Psychological review,

84(2), 191.

Bandura, A. (1977). Social learning theory.

(39)

Bandura, A. (2001). Social cognitive theory: An agentic perspective. Annual review of psychology, 52(1), 1-26.

Bayus, B. L. (2013). Crowdsourcing new product ideas over time: An analysis of the Dell IdeaStorm community. Management Science, 59(1), 226-244.

Becker, H. S., & Hughes, E. C. (1995). Making the grade: The academic side of college life. Transaction publishers.

Blumenstock, J. E. (2008, April). Size matters: word count as a measure of quality on wikipedia.

Proceedings of the 17th international conference on World Wide Web (pp. 1095-1096). ACM.

Brown, J., Broderick, A. J., & Lee, N. (2007). Word of mouth communication within online communities: Conceptualizing the online social network. Journal of interactive marketing, 21(3), 2-20.

Brown, J. D., Collins, R. L., & Schmidt, G. W. (1988). Self-esteem and direct versus indirect forms of self-enhancement. Journal of Personality and Social Psychology, 55(3), 445.

Bullinger, A. C., Neyer, A., Rass, M., & Moeslein, K. M. (2010). Community-based innovation contests: Where competition meets cooperation. Creativity and innovation management, 19(3), 290-303.

Chan, C. M. L., Bhandar, M., Oh, L. B., & Chan, H. C. (2004, January). Recognition and participation in a virtual community. System Sciences, 2004. Proceedings of the 37th Annual Hawaii International

(40)

Chang, H. H., & Chuang, S. S. (2011). Social capital and individual motivations on knowledge sharing: Participant involvement as a moderator. Information & management, 48(1), 9-18.

Chesbrough, H. (2006). Open innovation: a new paradigm for understanding industrial innovation. Open

innovation: Researching a new paradigm, 1-12.

Chiu, C., Hsu, M., & Wang, E. T. (2006). Understanding knowledge sharing in virtual communities: An integration of social capital and social cognitive theories. Decision support systems, 42(3), 1872-1888.

Chow, W. S., & Chan, L. S. (2008). Social network, social trust and shared goals in organizational knowledge sharing. Information & Management, 45(7), 458-465.

Dahlander, L., & Wallin, M. W. (2006). A man on the inside: Unlocking communities as complementary assets. Research Policy, 35(8), 1243-1259.

Dellarocas, C. (2005). Reputation mechanism design in online trading environments with pure moral hazard. Information Systems Research, 16(2), 209-230.

Dong, J. Q., & Wu, W. (2015). Business value of social media technologies: Evidence from online user innovation communities. The Journal of Strategic Information Systems. 24(2), 113-127.

Faraj, S., Jarvenpaa, S. L., & Majchrzak, A. (2011). Knowledge collaboration in online communities.

(41)

Füller, J. (2010). Refining virtual co-creation from a consumer perspective. California management

review, (52), 98-122.

Füller, J., Bartl, M., Ernst, H., & Mühlbacher, H. (2006). Community based innovation: How to integrate members of virtual communities into new product development. Electronic Commerce Research, 6(1), 57-73.

Gardner, W., Mulvey, E. P., & Shaw, E. C. (1995). Regression analyses of counts and rates: Poisson, overdispersed Poisson, and negative binomial models. Psychological bulletin, 118(3), 392.

Gebauer, J., Füller, J., & Pezzei, R. (2013). The dark and the bright side of co-creation: Triggers of member behavior in online innovation communities. Journal of Business Research, 66(9), 1516-1527.

Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revision of personal history. American

psychologist, 35(7), 603.

Hattie, J., & Timperley, H. (2007). The power of feedback. Review of educational research, 77(1), 81-112.

Heen, S., & Stone, D. (2014). Find the coaching in criticism. Harvard Business Review, 92(1-2), 108-+.

Hilbe, J. (2011). Negative binomial regression. Cambridge University Press.

(42)

Hu, M., & Liu, B. (2004, August). Mining and summarizing customer reviews. Proceedings of the tenth

ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 168-177). ACM.

Hutter, K., Hautz, J., Füller, J., Mueller, J., & Matzler, K. (2011). Communitition: The Tension between Competition and Collaboration in Community-Based Design Contests. Creativity and Innovation

Management, 20(1), 3-21.

Hsu, C. L., & Lin, J. C. C. (2008). Acceptance of blog usage: The roles of technology acceptance, social influence and knowledge sharing motivation. Information & Management, 45(1), 65-74.

Hsu, C. F., Khabiri, E., & Caverlee, J. (2009, August). Ranking comments on the social web.

Computational Science and Engineering, 2009. CSE'09. International Conference on (Vol. 4, pp. 90-97).

IEEE.

Hsu, M. H., Ju, T. L., Yen, C. H., & Chang, C. M. (2007). Knowledge sharing behavior in virtual communities: The relationship between trust, self-efficacy, and outcome expectations. International

Journal of Human-Computer Studies, 65(2), 153-169.

Ilgen, D., & Davis, C. (2000). Bearing bad news: Reactions to negative performance feedback. Applied

Psychology, 49(3), 550-565.

(43)

Jeppesen, L. B., & Molin, M. J. (2003). Consumers as co-developers: Learning and innovation outside the firm. Technology Analysis & Strategic Management, 15(3), 363-383.

Joinson, A. N. (2001). Self-disclosure in computer-mediated communication: The role of self‐awareness and visual anonymity. European journal of social psychology, 31(2), 177-192.

Jones, M. C., & Churchill, E. F. (2009). Conversations in developer communities: a preliminary analysis of the yahoo! pipes community. Proceedings of the fourth international conference on Communities and

technologies. ACM.

Jøsang, A., Ismail, R., & Boyd, C. (2007). A survey of trust and reputation systems for online service provision. Decision support systems, 43(2), 618-644.

Khabiri, E., Hsu, C. F., & Caverlee, J. (2009, March). Analyzing and Predicting Community Preference of Socially Generated Metadata: A Case Study on Comments in the Digg Community. ICWSM.

Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: a historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological bulletin, 119(2), 254.

(44)

Liu, N., & Carless, D. (2006). Peer feedback: the learning element of peer assessment. Teaching in

Higher education, 11(3), 279-290.

Magnusson, P. R. (2009). Exploring the Contributions of Involving Ordinary Users in Ideation of Technology Based Services*. Journal of Product Innovation Management, 26(5), 578-593.

Markus, H. (1977). Self-schemata and processing information about the self. Journal of personality and

social psychology, 35(2), 63.

Meyer, D., Hornik, K., & Feinerer, I. (2008). Text mining infrastructure in R. Journal of Statistical

Software, 25(5), 1-54.

Moon, J. Y., & Sproull, L. S. (2008). The role of feedback in managing the Internet-based volunteer work force. Information Systems Research, 19(4), 494-515.

Nambisan, S. (2002). Designing virtual customer environments for new product development: Toward a theory. Academy of Management Review, 27(3), 392-413.

Nambisan, S., & Baron, R. A. (2009). Virtual customer environments: testing a model of voluntary participation in value co-creation activities. Journal of product innovation management, 26(4), 388-406.

(45)

O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality &

Quantity, 41(5), 673-690.

Ocasio, W. (1997). TOWARDS AN ATTENTION-BASED VIEW OF THE FIRM WILLIAM OCASlO.

Psychology, 1, 403-404.

Ocasio, W. (2011). Attention to attention. Organization Science, 22(5), 1286-1296.

Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: a critical review of the literature and recommended remedies. Journal of applied

psychology, 88(5), 879.

Poetz, M. K., & Schreier, M. (2012). The value of crowdsourcing: can users really compete with professionals in generating new product ideas?. Journal of Product Innovation Management, 29(2), 245-256.

Reilly, R. R., Smither, J. W., & Vasilopoulos, N. L. (1996). A longitudinal study of upward feedback.

Personnel Psychology, 49(3), 599-612.

Resnick, P., Kuwabara, K., Zeckhauser, R., & Friedman, E. (2000). Reputation systems. Communications

of the ACM, 43(12), 45-48.

(46)

Rovai, A. P. (2002). Building sense of community at a distance. The International Review of Research in

Open and Distributed Learning, 3(1).

Ryan, R. M., & Deci, E. L. (2001). On happiness and human potentials: A review of research on hedonic and eudaimonic well-being. Annual review of psychology, 52(1), 141-166.

Schunk, D. H. (1990). Goal setting and self-efficacy during self-regulated learning. Educational

psychologist, 25(1), 71-86.

Shute, V. J. (2008). Focus on formative feedback. Review of educational research, 78(1), 153-189.

Tax, S. S., Colgate, M., & Bowen, D. E. (2006). How to prevent your customers from failing. MIT Sloan

management review, 47(3), 30-38.

Smither, J. W., Barry, S. R., & Reilly, R. R. (1989). An investigation of the validity of expert true score estimates in appraisal research. Journal of Applied Psychology, 74(1), 143.

Tesser, A., & Campbell, J. (1983). Self-definition and self-evaluation maintenance. Psychological

perspectives on the self, 2, 1-31.

Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness.

Administrative science quarterly, 35-67.

Van Der Schaar, M., & Zhang, S. Z. (2014). A dynamic model of certification and reputation.

(47)

Von Hippel, E. (2005). Democratizing innovation. MIT press.

Von Hippel, E. (2005). Democratizing innovation: The evolving phenomenon of user innovation. Journal

für Betriebswirtschaft, 55(1), 63-78.

Wang, S., & Noe, R. A. (2010). Knowledge sharing: A review and directions for future research. Human

Resource Management Review, 20(2), 115-131.

Wang, Y., & Fesenmaier, D. R. (2004). Towards understanding members’ general participation in and active contribution to an online travel community. Tourism management, 25(6), 709-722.

Wooten, J. O., & Ulrich, K. T. (2011). Idea generation and the role of feedback: Evidence from field experiments with innovation tournaments. Available at SSRN 1838733.

Wu, J. J., Chen, Y. H., & Chung, Y. S. (2010). Trust factors influencing virtual community members: A study of transaction communities. Journal of Business Research, 63(9), 1025-1032.

Wu, S. C., & Fang, W. (2010). The effect of consumer-to-consumer interactions on idea generation in virtual brand community relationships. Technovation, 30(11), 570-581.

(48)

Yoon, C., & Rolland, E. (2012). Knowledge-sharing in virtual communities: familiarity, anonymity and self-determination theory. Behaviour & Information Technology, 31(11), 1133-1143.

Zar, J. H. (1972). Significance testing of the Spearman rank correlation coefficient. Journal of the

(49)

Appendix 1

(50)
(51)

Referenties

GERELATEERDE DOCUMENTEN

Keywords: crowdsourcing; online user innovation communities; attention allocation; justification;

Now the main results have been evaluated, the research question can be answered: “Whether and how can focal users influence user feedback in OUICs?” The results of this study

Informational justice leads to popular ideas in combination with: the presence of the efficiency, effectiveness justification logic and the absence of the

With respect to the earlier mentioned benefits of process innovations, the theory of Schumpeter and based on the academic arguments from existing literature, it

Als je goed kijkt, vind je hier de meest bijzondere soorten en vegetatie.” Tijdens zijn onderzoek werd hij verrast hoe weinig dynamisch de rivierduintjes bleken waar ze een

The angular distribution of the density in the flow field downstream of a sonic nozzle is an issue of major interest in supersonic molecularbeam production, as

In order to explain the role of anthocyanin synthesis during early fruit development, we argued that anthocyanins in immature apple fruit protect the peel from

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of