![Sample our Information Science journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5940/TOC-Subject-Sample-2021-Communication-Studies.jpg"})
ABSTRACT
Digital platforms have transformed various industries, with profound changes witnessed in settings characterized by repeated, low value, and novel transactions, such as ride sharing, household services, and food delivery. Platform providers need to understand the factor(s) that suppliers consider before choosing to participate on their platforms. We argue that multi-level theorizing is necessary to explain the patterns of decision criteria that constitute the complex, yet boundedly-rational decision of platform participation. We draw on multiple theoretical perspectives which include predictors from firm-level strategic behavior, firm-level digital predictors, institutional predictors, and platform level, competition and contextual (environmental) predictors to model restaurant’s decision to participate on food delivery platforms, which are an exemplar of platforms that digitize transactions of ubiquitous and episodic nature. The population of 95,735 restaurants, serving a total of 135 different cuisines, located in the 37 largest cities of India, forms the dataset for developing our multi-level theory. Our decision tree induction methodology, which employs high levels of pruning, empowers us to discover context-specific rules that serve as credible approximations of the partial ordering of decision flows tacitly used by restaurants in deciding to participate on platforms. We identify six combinations of predictors, that is, decision rules, which are situated in at least four theoretical perspectives, to succinctly explain supplier participation on digital platforms. Finally, we abduct away from these specific decision rules to develop generalizable theoretical propositions. The decision trees, context-specific rules, general forms of the rules, and generalized propositions together form our multi-level theory for supplier participation on digital platforms. Our findings aid platform providers identify suppliers who can be the focus of efforts to increase platform participation, and help suppliers identify the participation status of competitors. For policymakers, our findings imply that incentives at the ends of the pricing spectrum can increase supplier participation on digital platforms.
Acknowledgements
We thank participants at seminars at the Asian School of Business, Chinese University of Hong Kong, Georgia State University, Indian School of Business, and University of Hawai’i at Mānoa for helpful comments on previous versions of this paper. Any errors that might remain are our own.
Supplemental Material
Online supplement for this article can be accessed at here
Notes
1. A substantive literature examines issues related to user generated content, such as reviews, ratings, and comments produced by consumers on platforms. Our study is not situated in this research conversation due to two reasons. First, production of user generated content usually does not impose an explicit participation cost on the consumers. However, the platform we examine has an explicit participation cost. Second, generation of content is not the primary purpose of the platform we study. In our study, restaurants are suppliers of the primary product – food, that is transacted via the platform.
2. A study by Yaraghi et al. [Citation96], which combined constructs across two levels, is an example of an exception.
3. Hypothetically, if India matches global peers in current averages for restaurant use (20%) and online delivery penetration (30%), it will witness more than 50 million online food delivery transactions a day. In 2019, UberEATs charges 0.23 cents per delivery from customers. Other platforms charge higher fees from either customers or restaurants.
4. Julius Pacuis introduced the concept of abduction in 1597 by translating the Aristotelian concept of apagoge [Citation68]. Peirce posited that abduction denotes the “only truly knowledge extending means of inferencing,” categorically distinct from deduction and induction [Citation59]. These ideas have been systematically received and adopted by various social sciences over the past several decades.
5. The definitions of suppliers and consumers are abstractions to increase applicability across a wide spectrum of platforms. For example, on a food delivery platform, restaurants are suppliers and individual users are consumers. Similarly, on a crowdfunding platform, entities who are seeking funding are consumers whereas those who provide funding are suppliers.
6. There are more than 500 restaurants containing the word “Sagar” in their name. Though these restaurants are not part of a chain, typically owners of many of these restaurants belong to the same community. Restaurants with the name “Shiv Sagar” form an institutional chain as they share common features such as offering a no frills, vegetarian, Indian cuisine experience for low prices. Number of Outlets, is therefore a contextually valid predictor.
7. The usefulness and high economic impact of product ratings and reviews for other users is undeniable (e.g., [Citation22]. However, since the producer knows his or her own quality completely, the impact of user reviews and ratings on the producer’s decision processes on a two-sided platform deserves further investigation. We investigate impact of user ratings on a restaurant’s decision to participate (or not participate) on the food delivery platform.
8. We collect data from this platform only at the end of November 2017 as there were three exogenous shocks which created extreme discontinuities, warranting data collection only after their effects had subsided. First, in November 2016, there was demonetization in India whereby large currency notes were declared void overnight. This was accompanied by a concerted policy push to encourage citizens and firms to adopt digital means for conducting business transactions. Second, a Supreme Court of India ruling in April 2017 banned restaurants from serving alcohol within 500 meters of national and state highways. As a consequence, numerous restaurants suffered large business losses. Later in August 2017, the Supreme Court clarified that this ruling does not apply within city limits. Third, in July 2017, a unified indirect tax called the Good and Services Tax (GST) was introduced by the Federal Government of India. This singular tax replaced existing multiple cascading taxes levied by various levels of government. Disruptions due to the introduction of GST were extensive since this increased the regulatory and administrative burden as all (documented and undocumented) parties in the value chain were dragged into the regulatory tax structure without any preparation. This increased the overall tax burden both on restaurants and consumers. This forced a change in the GST as later in November 2017, the government changed GST to reduce the tax rates to pre-GST rates. Our data collection is also timely considering subsequent changes in the food delivery ecosystem as a key platform was acquired in December 2017, while other platforms started aggressively increasing restaurant platform participation fees in subsequent time periods.
9. There exist other differences between the platform we choose to study, and other platforms present in the market. Although these inter-platform differences may plausibly drive participation decisions of restaurants on other platforms, they do not impact our study as none of these factors deter participation on the focal platform, which is an independent, non-exclusive decision.
10. Modern India is an amalgamation of multiple cultures with a shared history of several millennia. For example, the 2011 Linguistic Survey of India identified over 780 languages and 66 different scripts in use across the nation. Indian cuisines reflect this diversity and historical richness, with there being no pan-Indian cuisine. Indian cuisine can be broadly split into five categories – northern, southern, eastern, western, and northeastern, with the cuisine of each region reflecting its local produce, cultural diversity, and varied demographics. The restaurants in our dataset also reflect this diversity; they collectively serve 133 unique cuisines, out of which 35 are Indian and the remaining 98 are international cuisines.
11. We chose to limit our sample to restaurants from the top 7 metropolitan cities in India. Four of the older, mature metros included in our study are Mumbai (population approximately 20 million), Delhi (population approximately 25 million), Kolkata (population approximately 14 million) and Chennai (population approximately 8 million). Recently, the metros of Pune (population approximately 10 million), Bangalore (population approximately 10 million) and Hyderabad (population approximately 9 million) have seen rapid growth given the growth of the IT industry in India. Thus, the seven largest metros to which we restrict our sample encompass a total population of approximately 96 million people.
12. There were 35,815 (37%) missing values on the platform-level predictor – Rating. This is a key observation which deserves further exploration. However, this is likely to be the case when studying digital platforms as inactivity can be seen on most platforms (one outcome of the long tail phenomena/effect). Due to network effects-based incentives, platforms wish to display large number of participants to the other side of the market. Hence many platforms do not delete inactive participants. In our case, the platform does not reveal that a restaurant has exited the platform. We observe that a large number of the missing ratings in our population belong to inactive (closed) restaurants. Newly opened restaurants also do not have a rating for the first 3-months due to platform regulations. Finally, restaurants that do not have any user-generated reviews also do not have a rating. Based on our research design, objective of inductively developing theory by leveraging multi-level predictors, and since our data does not indicate the reason behind the lack of rating — that is, if the restaurant is closed, newly opened or lacks reviews, we excluded these data points from our analysis.
We conducted detailed ex post analysis on this sub-sample. First, we conducted a descriptive analysis. 12,679 of the restaurants without ratings were outside the top 7 metros, which is proportionate to the total number of restaurants outside the metros. This implies that in the smaller cities, the platform is arguably not used for reviews. There is no user generated content and reputation is not digitized or migrated to the digital medium. In other words, non-metro India is not digital, at least for restaurant reviews and delivery. This finding also validates our sampling criterion of only focusing on restaurants in the big 7 metros of India. A disproportionate number of restaurants (20,033) had low prices; these restaurants plausibly do not have enough reviews because their customers may not be technology savvy. Finally, nearly half of the restaurants (17,296) were indicated as participating on the platform. It is plausible that these restaurants were either shut down or newly opened as participating restaurants usually have ratings. Overall, we can infer that half of the restaurants without ratings did not have enough reviews, whereas the other half were either closed or newly opened. Second, we re-ran our main analysis for the entire population of restaurants in the 7 top metros, without including the platform-level predictor. Our main findings and resultant decision rules remain qualitatively unchanged.
13. The process to determine latitudes and longitudes of restaurants consisted of several steps. First, we programmed a script to use Google Maps API and find exact matches for the entire address of the restaurant. This yielded 65,065 matches for which latitudes and longitudes were extracted and used. The script was then rerun after deleting all words before the first, second and third commas in the address, yielding another 16,236, 1,863 and 388 coordinates respectively. Finally, we used Amazon Mechanical Turk to confirm the latitude and longitude coordinates for the 18,487 restaurants for which a non-exact match was retrieved by comparing the result from two different tools provided by Google and NASA. Only 937 coordinates met our strict condition that results from the two tools must be within 10 meters of one another. Thus, we were able to represent 66,002 restaurants in space.
14. Measures related to spatial concentration of competition can be created either as a count of all competitors within a specific distance to the focal firm, or as a distance-weighted measure (e.g., [Citation78, Citation82]). A criticism of count-based measures is that these ignore the presence of any firms that fall beyond a geographic boundary, irrespective of how close they lie to the boundary. Weighted density measures redress the problem of arbitrary boundaries. Although distance-weighted measures are considered nuanced, nevertheless, a count-based measure seems reasonable for our study due to the following logic. First, our qualitative analysis suggests that customers typically choose from restaurants within a walking or commutable geographic area. In other words, all competitors of a focal restaurant within a specific geographic area are relatively equal. Second, distance-weighted measures have been used in studies involving large distances between locations. The walkable or commutable distance pertinent for our study is much smaller as it is the distance that can be covered in a 10-minute walk/commute within a densely populated city, which is less than 1 kilometer. Hence the difference between a distance-weighted and a count-only measure would be insignificant at such small values. Finally, though our qualitative interviews suggested that a 0.5-kilometre radius is sufficient to capture all spatially concentrated restaurants, we chose 1 kilometer as our boundary out of an abundance of precaution.
15. Caution needs to be exercised when applying the C4.5 algorithm. Predictors with large number of (almost) continuous values can be mistakenly considered by the C4.5 algorithm to be informative. We guard against this potential concern by recoding continuous variables in our data into an appropriate number of (high/medium/low) bins. These choices are articulated clearly when we describe our predictors in the section describing our data.
16. We include a total of fourteen predictors across various levels to predict platform participation. It is not adequate to merely examine a decision point as our objective is to discover patterns of decision sequences, constituting the interdependencies between these fourteen predictors, not an individual decision or decision outcome alone. Our methodology of decision tree induction discovers combinations of predictors which are not likely to be known ex ante. In contrast, empirical methods such as probit regression methods require scholars to specifically construct and include interaction terms ex ante when predicting outcomes. Additionally, if scholars construct four-way interactions, all possible three-way interactions are required to be included in the regression models; leading to an explosion of interaction terms in the regression models. Furthermore, econometric approaches are not suited to our objective as our target is to study the decision journey and flow. Given these obstacles, we do not choose regression-based empirical models. Finally, the longest path in our decision tree incorporates an interaction of seven predictors – an emergent finding, not known ex ante. Thus, our choice of an inductive methodology is well-suited for knowledge discovery.
17. Note that the longest path in our trees is discovered in phase 2 of our analysis. This path, depicted as the leaves corresponding to number of outlets in Figure B1 of Online Supplemental Appendix B, corresponds to an interaction of seven predictors.
18. To improve readability of the trees, branches or leaves having similar outcomes (for example, low and medium ratings resulting in participation) were coalesced into a single branch or leaf respectively.
19. Studies find that hybrid strategic profiles result in lower performance as compared to purer strategic profiles [Citation7, Citation81]. This is due to mutual exclusivity of some strategies, vulnerability of intermediate strategic profiles, and increases in organizational complexity (see [Citation7] for an extensive review). Other studies demonstrate the viability of mixed strategic profiles by adopting contingent perspectives [Citation30, Citation81].
20. Prior research has also argued that multiple contingencies could also have a conflicting influence such that effect of a set of predictors is cancelled out by the influence of other predictors. However, we do not observe any conflicting contingencies.
21. The design of online platforms constantly evolves as platforms introduce new features to improve usability and phase out features that were not used in the past. Thus, given the constant evolution of platforms, constructing a panel data set is relatively difficult. Given the nature of the research objectives of our study, we leverage a cross-sectional data set to explore the influence of predictors across various levels of theory that guide platform participation.
Additional information
Funding
Notes on contributors
Abhishek Kathuria
Abhishek Kathuria ([email protected]) is an Assistant Professor of Information Systems at the Indian School of Business, India. He received his Ph.D. in Business Administration from the Goizueta Business School at Emory University. His research examines the business value of IT, focusing on innovation, digital platform strategies, and emerging economies. His work has been published in such journals as Journal of Management Information Systems, Information Systems Research, Communications of the Association for Information Systems, among others, and received multiple best paper nominations and awards at various academic conferences. Dr. Kathuria is an advisor to and co-founder of multiple startups and consults on business transformation, organizational turnarounds, and IT strategy with public and private corporations in Hong Kong, India, and the Middle East.
Prasanna P Karhade
Prasanna P Karhade ([email protected]) is an Assistant Professor of Information Technology Management, and Shidler College Faculty Fellow at the Shidler College of Business, University of Hawai’i at Mānoa. He holds a Ph.D. in Business Administration from the University of Illinois at Urbana-Champaign. He worked as a software engineer before embarking on an academic career. Dr. Karhade’s research interests include design of formal contracts for governing IT outsourcing relationships, IT governance, and the impact of IT on firm innovation. His work has been published in Information Systems Research and MIS Quarterly.
Benn R. Konsynski
Benn R. Konsynski ([email protected]; corresponding author) is the George S. Craft Distinguished University Professor of Information Systems and Operations Management at the Goizueta Business School at Emory University. He holds a Ph.D. in Computer Science from Purdue University. He has held faculty positions at the University of Arizona and Harvard Business School, and served as adviser and board member in public and private corporations. He was also named Baxter Research Fellow at Harvard, and Hewlett Fellow at The Carter Center. Dr. Konsynski specializes in issues of digital commerce and information technology in relationships across organizations, and has published in such diverse journals as Communications of the ACM, Harvard Business Review, Journal of Management Information Systems, MIS Quarterly, Decision Sciences, Information Systems Research, IEEE Transactions on Software Engineering, and many others.