Content area
Purpose
This study attempts to discover effective strategies for mobile commerce applications (apps) to grow their consumer base by releasing app strategic updates. Drawing on the landscape search model from strategy research, this study conceptualizes mobile app update strategy as three interdependent decisions, i.e. what business elements are changed in an app strategic update, how substantial the changes are and when strategic updates are released relative to the competitive environment.
Design/methodology/approach
Using a field data set of 1,500 strategic updates of seven rival apps in the mobile travel market, this study integrated fuzzy set qualitative comparative analysis (fsQCA) with econometric analysis to analyze how app strategic update decisions interdependently influence app performance.
Findings
This study identified three effective and one ineffective mobile app update strategies from the mixed-method analysis, which verified the complex interdependency of app strategic update decisions. A general takeaway from these strategies is that a complex strategy problem on the mobile platform must be solved with respect to the constraints and capabilities of mobile technology.
Originality/value
This study moves beyond a linear view of the relationship between app update frequency and app performance and provides a holistic view of how and why app strategic update decisions mutually influence one another in their impact on app performance. This work makes contributions by identifying interdependency as a conceptual bridge between strategy and mobile app literature and developing an empirically testable version of the landscape search model.
1. Introduction
The pervasive adoption of mobile devices and the wide availability of the mobile web have turned mobile applications (apps) into a preferred shopping channel of consumers (Leon, 2018). In the 2019 holiday shopping season, consumers made more purchases on mobile phones than on desktops (Analytics, 2019). This consumer trend has turned mobile commerce apps into a new battleground for companies to gain a competitive advantage. A company must not only launch a commerce app to attract mobile customers, but also continually update the app to grow its mobile customer base amid similar attempts of rival apps (Kajanan et al., 2012).
App updates have been identified as an important strategic instrument for companies to grow their mobile customer base (Chen et al., 2021; Sun et al., 2022; Huang et al., 2017; Foerderer and Heinzl, 2017). Unlike non-feature updates such as bug fixes, hotfixes and patches, app strategic updates that this paper focuses on refer to the updates of introducing new business offerings or technical functionalities to an app (Foerderer et al., 2018; Comino et al., 2016). For example, Expedia accelerated its mobile customer base growth by releasing a series of app strategic updates such as introducing rental cars and adding “Things to do” options. These app updates are strategic for companies, because they are competitive moves taken by companies with the aim of promoting app downloads in hypercompetitive mobile app market (Chen et al., 2021; Foerderer and Heinzl, 2017). When continued strategic updates become a common practice in a cohort of rival apps, a deliberate app update strategy is required for a company to harness the effect of strategic updates. Prior research has identified key decisions entailed in app update strategy such as what elements in an app should be strategically updated (Comino et al., 2016), how substantial a strategic update should be (Gutt et al., 2019) and when should a strategic update be released with respect to previous updates of a focal app (Kajanan et al., 2012) and concurrent updates of rival apps (Tian et al., 2020).
Although prior research urges app developers to weigh several key decisions for app update strategy, it has not provided theoretical insights regarding how the strategic decisions can complement or compromise one another’s impact on app performance. Our study seeks to extend the mobile app research literature by embracing the interdependency among app strategic update decisions. In addition to guiding app update practice, our focus on the interdependency among strategic update decisions is valuable because it can reveal theoretical insights specific to the “mobile” and “strategic” nature of app updates. Strategy has been recognized as a complex problem composed of interdependent decisions (Ghemawat and Levinthal, 2008). App strategic updates, as strategic moves on mobile devices, are subject to a multitude of possibilities and constraints affected by mobile technology characteristics such as a small screen, limited storage and high portability (Pitt et al., 2011). For example, a small screen constrain the amount of business offerings that can be simultaneously introduced to customers (Ghose et al., 2012), while high portability enables location-based product recommendations to ease customers' product search on a small screen (Pitt et al., 2011). It can take a series of interdependent app strategic updates to balance the tradeoff between the possibilities and constraints.
We draw on the landscape search model from strategy research (Levinthal, 1997; Ghemawat and Levinthal, 2008) to develop an interdependency-centered, empirically testable research model of mobile app update strategy. The landscape search model compares performance outcomes of a combination of strategic decisions to a rugged landscape because one decision can cause disproportionally small or large performance changes due to its interdependency with other decisions. Accordingly, we analyze the combination patterns of the key strategic decisions entailed in app update strategy rather than the independent effect of each decision. By comparing performance outcomes from different combination patterns of the interdependent decisions, we seek to identify effective or ineffective configurations of mobile app update strategies.
Our research model was tested with a field data set of 1,500 app strategic updates released by seven rival apps in the mobile travel market during a 36-month period. We employed the fuzzy set qualitative comparative analysis (fsQCA) to discover superior or inferior combination patterns of the interdependent app strategic update decisions. Then, an econometric analysis was used to confirm the positive/negative impact of combinations of interdependent app strategic update decisions on app performance. We linked our findings to real-world cases and developed generalizable app update strategies. In particular, three effective and one ineffective mobile app update strategies were identified, which reveals how young apps or mature apps combine different changes of business elements under high or low competitive environments to obtain superior or inferior app performance. These findings demonstrate the complex interdependency among the strategic decisions of mobile app updates. Our findings also highlight the need to solve such a complex strategy problem of mobile app updates within the capabilities and constraints of mobile technology. For example, service innovations can build on the portability of mobile devices to provide location-based consumer experiences. When service-oriented app updates are combined with product-related updates, an app achieves superior performance.
With these findings, this study contributes to the literature in three ways. First, our study identifies interdependency as a conceptual bridge that links the grand theory of interdependent strategic decisions and phenomenon-specific observation of the interdependency of mobile technology characteristics. The second contribution of our study is to develop an empirically testable version of the landscape search model, which paves the way for extending computer simulations of the landscape search models to empirical analyses of digital trails of business strategy. Third, this study contributes to the mobile app research literature by moving beyond a linear view of the relationship between app update frequency and app performance. Compared with prior studies that mainly emphasize the frequency of app updates and their linear effects on app performance (e.g. Comino et al., 2016; Gutt et al., 2019; Tian et al., 2020), our landscape search model helps us open the black-box of decision-making involved in app updates and reveals the nonlinear combination patterns of the strategic update decisions.
2. Literature review and theoretical foundation
2.1 Prior mobile app update research
In prior mobile app research, app updates have been examined in two primary ways. One stream of models includes app updates as a control variable (e.g. Ghose and Han, 2014; Lee and Raghu, 2014). Measures of app updates focus on quantity such as the number of update release in a time period rather than quality of changes entailed in an update. As these studies only look at quantity of updates rather than content of an update, their findings have limited implications to how to formulate each update as a strategic move.
A second stream of mobile app research models recognize the strategic role of app updates and include app strategic updates as a core predictor of app performance (Kajanan et al., 2012; Yin et al., 2014; Comino et al., 2016; Gutt et al., 2019; Tian et al., 2020). These models seek to identify specific attributes of strategic updates that can bring superior app performance. Three groups of attributes are explored.
The first group of strategy attributes is concerned with types of changes included in a strategic update. For example, Comino et al. (2016) find that the positive effect of app strategic updates on app downloads is stronger in iTunes store where strategic updates are reviewed than in Google Play where a review process is lacking. They conjecture that the review process in iTunes store nudges developers to include substantive changes in a strategic update. Tian et al. (2020) distinguish between strategic updates that introduce a new category of products or services (namely app core innovation) and those that add supporting functionalities to existing products or services (namely app support innovation).
Prior research has also explored a second group of strategic update attributes that reflect the magnitude of changes in a strategic update. For example, in Comino et al. (2016) major updates are defined as those with significant jumps in functionalities and minor updates are those that focus on fixing bugs. It shows that major updates stimulate app downloads more effectively than minor updates. Similarly, Gutt et al. (2019) find that major updates can enhance an app’s customer rating while minor updates' impact on customer rating vary from slightly positive to negative depending on an app’s previous ratings.
The third group of strategic attributes of app updates center on an app’s previous strategic updates and its rival apps' strategic updates. For example, Kajanan et al. (2012) find that inferior performance of an app’s previous strategic updates in promoting app downloads can lower the app’s current chance to enter and remain in the 240 most downloaded app list of iTunes store. Meanwhile, Tian et al. (2020) show that when a focal app’s updates are perceived as imitation of rival apps, customer reviews tend to deteriorate.
Insights from prior research taken together indicate that app developers must weigh several key strategic decisions regarding types of changes, magnitude of changes and timing and competitive context of changes entailed in app strategic updates. While prior research has provided partial understanding about the importance of each strategic decision, there lacks a holistic view of how these key decisions can mutually influence one another’s impact on app performance. Nevertheless, a few two-way interaction terms in prior research models have hinted at the existence of interdependency among strategic update decisions (e.g. Gutt et al., 2019; Tian et al., 2020).
More importantly, the existence of interdependency of app strategic updates has been indicated by prior discussions of the “mobile” and “strategic” nature of app updates. First, technological characteristics of mobile devices, such as a small screen, limited storage and high portability, bring a multitude of possibilities and constraints to mobile app strategic updates (Pitt et al., 2011; Ghose et al., 2012). A series of interdependent app strategic updates can balance the tradeoff between the possibilities and the constraints. For example, an app strategic update that introduces a variety of new product categories should come with another app strategic update that enhances product searches in the small screen. Second, mobile app strategic updates, as strategic moves to attract new customers, reveal the intrinsic nature of a strategy problem. Prior research has established strategy as a complex problem composed of many interdependent decisions (Ghemawat and Levinthal, 2008). Mobile app strategic updates are likely to be implicitly or explicitly managed as interdependent decisions. A research focus on interdependency among app strategic update decisions can provide new insights specific to the mobile and strategic nature of app updates.
A variance model with independent and dependent variables is unlikely to produce a holistic view of the interdependency among app strategic update decisions because the interdependency can be more complex than two-way or three-way interaction effects in a variance model (Tang and He, 2021). We, therefore, choose a modeling approach suitable for analyzing complex interdependency in strategy – the landscape search model explained below.
2.2 The landscape search model
The landscape search model originated in biology research on interdependent changes within an organism as related to the organism’s fitness in an environment (Kauffman, 1993). Based on the metaphor of a mountainous landscape, the landscape model uses the altitude of a point in the landscape to represent the performance contribution of a set of interdependent changes in an organism. The search for an advantageous combination of interdependent changes is compared to finding high peaks in the mountainous landscape (Kauffman, 1993).
The landscape search model is extended from biology to strategy research because strategy, like an organism, is a complex problem composed of many interdependent decisions (Levinthal, 1997; Ghemawat and Levinthal, 2008; Baumann and Siggelkow, 2013). The metaphor of a rugged landscape can illustrate that the performance contribution of interdependent decisions involved in a complex strategy problem differs from the sum of the performance impact of each decision. A company’s strategic moves toward superior performance are compared to searching in a mountainous landscape for higher peaks (Levinthal and Warglien, 1999; Tanriverdi et al., 2010).
Prior applications of the landscape search model have reduced a complex strategy problem to three interdependent decisions, which can be mapped to the three key decisions entailed in app update strategy. The first decision is to identify core business elements that contribute to business performance (Siggelkow, 2002). This decision is important because the core business elements and interdependency among them determine the ruggedness or mountainous contours of the strategy solution space (Ghemawat and Levinthal, 2008; Baumann and Siggelkow, 2013). Furthermore, core business elements can vary across companies and competition environments. Thus, a company must identify the core business elements in its own strategy (Siggelkow, 2002). The decision regarding what types of changes to include in an app strategic update corresponds to the landscape search decision to characterize the contour of the strategy solution space. Using the language of the landscape search model, we can specify the first key decision in an app update strategy as what core business elements should be changed in app strategic updates?
While the first decision characterizes the contour of the landscape, the second decision is to find an effective search approach to reach high peaks in the landscape (i.e. obtain superior performance). Prior landscape search models have identified two typical search approaches: local search and distant search. Local search produces incremental changes of a company’s core business elements by exploiting its expertise or knowledge (Levinthal and Warglien, 1999; Fleming, 2001). The local search approach quickly locks a company in the first peak encountered, but unlikely reaches the highest peak in the rugged landscape (Levinthal, 1997). Distant search leads to radical changes in core business elements far removed from a company’s current model of operation by exploring new knowledge (Fleming, 2001). In a rugged landscape, distant search allows a company to jump out of a low peak, but increases the uncertainty of performance outcomes (Levinthal, 1997). Given the tradeoff between the two search approaches, a company must choose the right search approach according to the contour of the strategy solution space and their current position in the landscape. Likewise, for developers of mobile apps, they must decide how substantially core business elements should be changed in an app strategic update.
Third, a company must choose the right timing for strategic moves (Chao and Kavadias, 2008). The contour of the strategy solution space and the effectiveness of a search approach depend on the maturity of a focal company’s strategy and the strategic moves of its competitors (Gavetti and Rivkin, 2007). The unexplored area in the strategy solution space becomes smaller after a company has taken a greater number of strategic moves. The core business elements and appropriate search approach can change in the reduced solution space. In addition, a greater number of strategic moves by competitors is more likely to reshape a focal company’s strategy solution space and derail its search approach (Levinthal and Warglien, 1999; Tanriverdi et al., 2010). Insights from the landscape search model help us to define the third app strategic update decision as when should an app strategic update be released in relation to the maturity of the focal app and its rival apps' updates?
3. Research model
Prior landscape model research has mainly used context-independent computer simulations to test the complex interdependency among strategic decisions. We develop an empirically testable research model by mapping the three app strategic update decisions presented above to concepts measurable with field data.
Regarding core business elements involved in mobile app strategic updates, prior e-commerce research has identified two distinguishable and measurable business elements in commerce apps: products and services (Tian et al., 2020). Products refer to tangible or intangible goods sold in an app (Mocker et al., 2014). Services are free assistance, such as product recommendations, offered by an app to facilitate customer purchase of its products (Cenfetelli et al., 2008). These two business elements are the primary reasons for consumers to use a commerce app as a shopping channel and are, therefore, the prominent and observable manifestations of core elements involved in app updates. The technological characteristics of mobile devices can bring unique opportunities or challenges for product or service offerings in an app (Ghose et al., 2012). Therefore, a company is likely to strategically update product and service offerings as part of its mobile-specific strategy.
An app strategic update can present an incremental change (i.e. local search) or a radical change (i.e. distance search) in products or services. Previous research has empirically measured the two search approaches by the extent of managers' familiarity with the knowledge used in a search approach (Fleming, 2001). Incremental change corresponds to the change of elements by using familiar knowledge (e.g. refining previous elements), while radical change refers to changing elements with new knowledge (e.g. developing a new element) (Fleming, 2001). Following prior research, we define the magnitude of core element change by a company’s familiarity with the knowledge required by the change. Specifically, incremental change refers to the change of product or service by using familiar knowledge, such as refining a payment function and increasing the inventory of pre-book tickets. Radical change refers to the change of a product or service by developing new knowledge, such as adding overseas hotels as a new product category and launching a new map-based search function.
The timing of an app strategic update is defined by the focal app’s maturity and rival apps' strategic updates. Prior app update research shows that app age moderates the impact of app version changes on app performance (Kajanan et al., 2012). Meanwhile, rival apps' updates also influence the effectiveness of the focal app’s update strategy (Tiwana, 2015). In the mobile platform, where strategic updates of all apps are visible to customers and competitors, changes of core business elements in a focal app often trigger a rival apps' subsequent changes (Tiwana, 2015). The competitive interaction may redefine the payoff of each company’s app strategic updates.
To embrace the causal complexity among the app strategic update decisions, our research model focuses on combination patterns of the measurable concepts that embody the app strategic update decisions. By comparing app performance outcomes resulting from different combination patterns, we can infer effective or ineffective mobile app update strategies. Yet, unlike landscape search simulation models that implement different combinations of model settings as experimental conditions, our study detects combination patterns of the concepts embedded in a series of app strategic updates in the real world. Table 1 summaries the interdependent strategic decisions, the measurable concepts and the supporting references.
4. Research method
4.1 Research context and data collection
We chose the mobile-based travel market in China as our research context for two reasons. First, travel products and services include a wide range of online and offline options (e.g. tours can be booked online but must be consumed offline), which drive companies to explore capabilities and constraints of mobile technology in integrating consumers' online and offline experience via a series of mobile app strategic updates (Tan et al., 2017). Second, a cohort of online travel companies in China began to invest heavily in their mobile commerce apps in 2013 (Ctrip Annual Report, 2013; Elong Annual Report, 2013). They jointly produced a period of app update dynamics typical of rivalry companies' strategic moves to attract consumers in an emerging mobile-based market.
Within the mobile-based travel market, our data collection focused on seven prominent travel apps – Ctrip, Elong, Qunar, Tuniu, Fliggy, Lvmama and Tongcheng – between May 2013 and April 2016 (i.e. 36 months). These seven apps accounted for over 75% of the entire mobile-based travel market in China during 2013 and 2016 and were consistently ranked among the top ten most popular apps in the China online travel industry during the same period. Moreover, as the dominant players in mobile-based travel market, these seven apps compete over the same pool of prospective customers, which ensures a rivalry relationship among them (Chen, 1996).
We used apps' release notes as the primary data source because they provided comparable descriptions about strategic updates of different apps. Release notes of different apps are written in a similar style. They describe what app features are changed and how substantial the changes are in a new release. We collected 407 release notes for our sampled apps' strategic updates from shouji.com.cn, which is one of the largest app stores in China. Besides release notes, we collected news articles regarding the seven mobile apps as supplementary data. Online travel companies routinely hold press conferences to introduce their new app versions, which can generate news reports about mobile app strategic updates. A total of 440 news articles related to the sampled apps were collected.
Data regarding app performance was collected from Ctcnn.com. This news portal provides monthly statistics about the number of downloads of top ten travel apps in major app stores in China. The statistics only count the number of first-time user downloads, but not the number of existing users' download of app updates (Foerderer and Heinzl, 2017). Therefore, our app performance data captured an app’s customer base growth. The reliability of our data source has been reviewed and endorsed by authoritative institutions including China e-business research center.
Our final sample includes a total of 252 app-month observations across seven competing apps and 36 months. Such monthly observations allow us to capture different updating strategies of mobile apps at different maturity level in hypercompetitive environment, as well as the performance outcome of these updating strategies. Choosing a month as the time unit also enables us to compare the empirical results of our research model with prior app update studies that also used the same time unit (e.g. Comino et al., 2016; Gutt et al., 2019; Tian et al., 2020).
4.2 Measurement
We employed content analysis to extract quantitative measures of the core business elements involved in app strategic updates (i.e. products and services) and the magnitude of changes (i.e. incremental and radical) from app release notes and news articles. The unit of analysis was an app strategic update announced in an app release note or news article. For example, Tuniu app on January 2015 announced three strategic updates—introducing train tickets, upgrading online chat functionality and improving personalized tour bundling functionality.
The content analysis was performed by two independent coders. They were given a coding scheme that included the definitions of products, services, incremental changes and radical changes as shown in Table 1. The coding scheme also provided examples of products and services in travel apps to guide the coders' identification of these two business elements. The coding scheme identifies the magnitude of changes with terms typical of incremental changes such as “upgrading,” “adding more,” and “expanding,” and those typical of radical changes such as “adding new” and “newly introduced.” The coders were asked to categorize each app strategic update to either a product-related or service-related update, but not both. Meanwhile, the coders categorized each strategic update as either incremental or radical, but not both. Discrepancy in coding results were resolved by a discussion among the coders and authors. To assess our coding reliability, a third coder recoded 10% of app strategic update notes. Based on the procedure in Perreault and Leigh (1989), the reliability index of our coding procedure was estimated as 0.90, which is higher than the recommended level of 0.70 (Chi et al., 2010).
The content analysis yielded quantitative measures of 1,500 strategic updates deployed by the seven sampled apps between May 2013 and April 2016. Each strategic update was coded as one of the four possible types: product radical change, product incremental change, service radical change and service incremental change. The monthly total number of strategic updates belonging to each type was used as the measure of four concepts regarding mobile app update strategy. For example, on December 2013, Qunar’s app strategic updates were measured as one product radical change (i.e. introducing airport-transfers as a new product category), four product incremental changes (i.e. expanding hotel inventory, increasing the taxi fleet for pre-booking, expanding the service area of shuttle buses and increasing the number of car rental suppliers), one service radical change (i.e. adding a new search keyword prompt feature) and one service incremental change (i.e. upgrading the existing search functionality).
The other two concepts regarding app strategic updates were measured by subjective data. Focal app’s maturity was measured by the number of months since the focal app was first released (Tiwana, 2015). Rival apps' updates were measured by the number of strategic updates announced by the other six rival apps in a given period (Chen, 1996).
The outcome of app updating strategy was measured by the growth rate of the monthly downloads of a focal app. The growth rate of app downloads reflects a company’s success in growing its mobile customer base, which is a predominant concern for companies competing in an emerging mobile market (Huang et al., 2017). We adopted the formula in Huang et al. (2017) to calculate the monthly growth rate of app downloads as: (downloads of a focal app in month t)/(the total downloads of the focal app from its initial release to month t-1).
Given that using the growth rate of app downloads as the outcome of app updating strategy may cause endogenous problem, we combined fsQCA with econometric analysis to analyze our research model. Several additional analyses were also performed to test the results robustness. The details of our analysis approach were provided in the below.
4.3 Analysis approach
We employed fsQCA to infer the complex interdependency among app strategic update decisions because of its following three advantages. First, fsQCA is a set-theoretic approach to examine the relationship between combinations of causal conditions and an outcome (Fiss, 2011; Tang and He, 2021). It assumes that an outcome results from interaction rather than independent effects of causal conditions (Misangyi et al., 2017). Second, fsQCA allows more than one combination of causal conditions for a given outcome. Third, conditions causally related in one combination may be unrelated or even inversely related in another (Zhan et al., 2023). In addition, given that the longitudinal nature of our dataset, we used the approach proposed by García-Castro and Ariño (2016) to evaluate whether the configuration results of the fsQCA vary over time across apps.
Moreover, following the advice of prior fsQCA research (Fiss et al., 2013), we supplemented the fsQCA with an econometric analysis. The econometric analysis not only helps us to evaluate the statistical significance of effective or ineffective combinations of causal conditions, but also reduces concerns of endogenous problem with econometric model specifications.
Model specifications for fsQCA. The six concepts embodying app strategic update decisions (i.e. product radical changes, product incremental changes, service radical changes, service incremental changes, focal app’s maturity and rival apps' updates) were included as causal conditions in fsQCA, while the growth rate of an app’s monthly downloads was the outcome. The measures of the causal conditions and outcome were first calibrated into membership scores ranging from 0 to 1 according to a threshold for full membership, a threshold for full non-membership and a threshold for crossover or the maximum fuzziness in membership (Ragin, 2008).
To set the thresholds for calibration, we followed the common approach in fsQCA studies to consider the distribution of the data sample and the practical meaning of the threshold values (e.g. Fiss, 2011; Misangyi and Acharya, 2014). For the measures of app performance outcome, focal app’s maturity and rival apps' updates, we followed the calibration approach of performance outcome and industry competitiveness in prior fsQCA analyses (e.g. Fiss, 2011; Misangyi and Acharya, 2014; Campbell et al., 2016) to set calibration thresholds. This approach employed the upper- and lower-quartile scores of the industry, respectively, as the full membership and full non-membership thresholds. In our study high app performance thresholds for full membership and full non-membership were set as 19.6% and 5.8%, respectively, which correspond to the 75th and 25th percentiles of the monthly download growth rate in our sample. The crossover threshold was set as the mean of the full membership and the full non-membership thresholds (i.e. 12.7% growth rate, close to the 50th percentile). The membership thresholds of low app performance were the reverse of those of the high app performance: 5.8% as full membership threshold and 19.6% as full non-membership threshold. The thresholds for calibrating focal app’s maturity and rival apps' updates were also based on the 75th and 25th percentiles of their respective distributions in our sample. Specifically, 55 months was the threshold for the full membership of focal app’s maturity, 46 as the crossover (the median in the sample, close to the mean) and 37 for the full non-membership. The full membership threshold for rival apps' updates was 52 updates (75th percentile in the sample), 39 as the crossover (the median in the sample, close to the mean) and 25 as the full non-membership threshold (25th percentile).
For the measures of product radical changes, product incremental changes, service radical changes and service incremental changes, we did not find existing thresholds to apply. We, therefore, followed prior fsQCA studies (e.g. Greckhamer, 2016; Misangyi et al., 2017) to set calibration thresholds based on sample distribution. To ensure practically meaningful threshold values, we set the full membership threshold at the 90th percentile and the absence of the change as the full non-membership threshold. Specifically, the full membership thresholds for product radical changes and product incremental changes are both set at 2 (i.e. the 90th percentile for these two measures have the same value—2 counts of change), with 0 as the full non-membership threshold and 1 as the crossover point (i.e. the mean of the full membership and full non-membership thresholds). Similarly, the full membership thresholds for service radical changes and service incremental changes were based on the 90th percentile of their measures, which had the value of 4 (i.e. 4 counts of change in a month). The full non-membership threshold was 0, and the crossover point of the two causal conditions was 2 (i.e. the mean of the full membership and full non-membership thresholds).
After the calibration process, the truth table with 26 rows listed all possible combinations of the six causal conditions. We followed prior fsQCA studies to reduce the number of combinations for subsequent analysis (Fiss, 2011). First, we set the minimum number of cases required in a combination as three. Using three cases as the frequency cutoff retained 88% observations of our sample. Second, the truth table calculated the consistency for cases in each combination to produce a given outcome. The minimum acceptable consistency score was set at 0.80 (Ragin, 2008). Third, we set 0.65 as the cutoff for proportional reduction in inconsistency (PRI) to avoid the simultaneous combinations that explain both the presence and absence of high performance or low performance (Greckhamer, 2016). Following the common practice in fsQCA studies, we used parsimonious solutions and intermediate solutions as fsQCA results (Fiss, 2011).
Model specifications for econometric analysis. Using the same set of 252 app-month instances, we constructed econometric models to compare the combination effects of app strategic updates decisions on app performance with the independent effects of decisions. To correct the selection bias and the possibility for mobile app update decisions to be endogenous, we employed a two-stage Heckman (1979) estimation procedure to perform the econometric analysis. As shown in Equation (1), we first used a probit model to estimate the likelihood of an app update. The dependent variable (i.e. ) in the first-stage selection model is a dummy variable, with 1 representing that a focal app i makes updates in a given month m and 0 otherwise. According to prior research, we introduced the number of sibling apps that the same company simultaneously develops as the exogenous variable to predict the likelihood of an app update (Zhou et al., 2018). If a focal app has a large number of sibling apps, its company will have limited resources and attention to update the focal app (Tsai, 2002). Thus, the number of sibling apps is an exogenous variable that affects a company’s decision on whether to update the focal app but does not directly affect the focal app’s performance (Zhou et al., 2018). In addition, the first-stage selection model also included several control variables () that potentially influence the update of focal app, including focal app’s maturity measuring with the number of months since focal app was first released, rival apps' updates measuring with the number of strategic updates announced by the other six rival apps in a given period and download base of prior month measuring with the number of the total downloads of the focal app from its initial release to month t-1. Moreover, given that the investment relationships between our sampled apps' companies may influence app update strategies, a dummy variable that represents whether the focal app’s company has established investment relationship with rival apps' companies was included as a control variable to address potential confounding effects. We collected information about investment relationship among our sampled apps' companies from their official websites and news articles.
As shown in Equation (2), we further built a second-stage outcome model to test the impacts of update strategy on app performance. The dependent variable () here is the outcome variable of the fsQCA model, i.e. the growth rate of the monthly downloads of a focal app i in a given month m. As for independent variables, we followed the method in Fiss et al. (2013) to calculate the membership score of an app-month instance in each effective or ineffective combination of causal conditions identified in the fsQCA. Thus, the six membership scores indicated the degree of inclusion of each app-month instance in a combination, which were included as independent variables (i.e. ). The inverse Mills ratio generated from the first-stage probit regression was included in the second-stage regression analysis to adjust for the potential selection bias. The control variables () of the second-stage model not only contain variables in (i.e. focal app’s maturity, rival apps' updates, download base of the prior month and a dummy variable indicating investment relationship), but also include the number of app versions and promotion activity. The number of app versions was a count of new versions released in a month, which is a common measure of app updates in prior literature (e.g. Lee and Raghu, 2014). Promotion activity was measured as a dummy variable to indicate whether an app announced a price cut in its release notes and related news articles in a month. Given the long panel structure of our dataset (i.e. many time periods and few individuals), app dummy variables and a calendar time trend variable were included as control variables.
First-stage selection model:
Second-stage outcome model:
The second-stage outcome model described above could indicate whether effective or ineffective combinations of causal conditions have a positive or negative impact on app performance, which is referred as the combination effects model hereafter. However, it did not rule out the possibility for each causal condition to independently influence app performance. To address this limitation, we developed a secondary regression model (i.e. independent effects model) that included the raw measures of the causal conditions involved in each app-month instance as independent variables, including the number of product radical changes, product incremental changes, service radical changes and service incremental changes. Our assumption that app updates do not independently affect app performance would be supported if the combination effects model has more significant coefficients and a higher R2 than the independent effects model.
The estimation of the combination effects model and independent effects model used both feasible generalized least squares (FGLS) and ordinary least squares (OLS) with panel-corrected standard errors (OLS-PCSE). These estimation techniques were chosen for their advantage in handling potential heteroscedasticity and autocorrelation problems in our time-series, cross-section dataset (Beck and Katz, 1995). Following prior applications of the estimation techniques (e.g. Han et al., 2011), our FGLS estimation models were specified to correct for heteroscedasticity and autocorrelation only, while the OLS-PCSE estimation models were specified to correct for heteroscedasticity, contemporaneous correlation and autocorrelation. In both FGLS and OLS-PCSE estimations, we corrected for autocorrelation by using panel-specific first-order autocorrelation (PSAR1) command.
5. Results
5.1 fsQCA results and robustness tests
The descriptive statistics and correlations of the measurements included in fsQCA are summarized in Table 2. Necessary conditions analysis of fsQCA suggest that none of six casual conditions about mobile app update strategy is necessary for high or low app performance.
Table 3 and Table 4 present our fsQCA results, which, respectively, display three high-performing and three low-performing combination patterns of app strategic decisions. In these two tables, black circles (“”) represent the presence of a condition; crossed-out circles (“”) represent the absence of a condition; and, blank spaces indicate “don’t care,” meaning that the presence or absence of a condition is not causally related to the outcome. Large circles represent core conditions, which appear in both parsimonious solutions and intermediate solutions. Small circles represent peripheral conditions, which are causal conditions that appear only in intermediate solutions.
Tables 3 and 4 also report the measures of consistency and coverage for individual combinations and for the collection of combinations as a whole. Consistency evaluates the degree to which cases reliably produce the outcome. The consistency scores for individual combinations and the collection of combinations are all above the minimum acceptable value of 0.75 (Fiss, 2011). Coverage indicates the degree to which a combination or a collection of combinations accounts for the outcome. Raw coverage takes into account a case’s membership in multiple combinations, while unique coverage explains a case’s non-redundant membership in a combination not covered by other combinations (Ragin, 2008). Overall solution coverage is analogous to R-square in regression analysis (Ragin, 2008).
Several additional analyses were performed to test the robustness of fsQCA results. We varied the fully-in thresholds and fully-out thresholds of all measures in fsQCA to perform a sensitivity analysis. For example, we adjusted the fully-in and fully-out thresholds of focal app’s maturity, respectively, to 70%th (or 80%th) percentile and 20%th (or 30%th) percentile of the sample distribution. Only slight changes in peripheral conditions were observed while our core conditions remained unchanged. This sensitivity analysis alleviates the concern that our fsQCA results are artifacts of the specification of thresholds. Moreover, in order to evaluate whether our fsQCA results are robust over time and across apps, we followed the approach of García-Castro and Ariño (2016) to calculate between consistency (BECONS) and within consistency (WINCONS). BECONS measures the cross-sectional consistency for each time unit in the panel, while WINCONS measures the longitudinal consistency for each firm over time. This additional analysis results also indicate our fsQCA results are stable across apps over time (García-Castro and Ariño, 2016).
5.2 Econometric analysis results and robustness tests
Table 5 presents the results of the first-stage selection model. As we expected, the number of sibling apps has a negative impact on the likelihood of the focal app’s update. In addition, the focal app’s maturity and rival apps' updates will prompt a company to update its app, while the increase in the download base of the prior month will impede a company to update its app. These results are consistent with the results of prior research (Huang et al., 2017). We also found that the investment relationship has no effect on the likelihood of updates.
As shown in Table 6 that lists the results of second-stage outcome model, the results of FGLS and OLS-PCSE consistently show that all three high-performance combinations have statistically positive effects on app performance and the low-performance combination LP1 has a negative and marginally significant effect on app performance and the effects of LP2 and LP3 are not significant. Moreover, the OLS-PCSE results show that compared with the R2 of the independent effects model (i.e. 13.34%), the R2 values of the combination effects model increased to 19.85%. This indicates the explanatory power of the combination effects of interdependent app strategic update decisions.
To further assess the robustness of our econometric analysis results, we also estimated the combination and independent effects models with fixed effect estimator or random effect estimator. As shown in Table 7, consistent results from FGLS and OLS-PCSE, fixed effect and random effect models indicate the robustness of our results against different estimation procedures.
To check whether there are any other endogeneity issues, we included instrument variables in our model to perform the Durbin–Wu–Hausman (DWH) test. Following prior studies (Gu et al., 2012), we chose one-month lagged independent variables (i.e. six membership values of an app’s updating strategy in the previous month in three effective or three ineffective combinations) as the instrument variables. The p-values of both the Durbin test (0.24) and the Wu–Hausman test (0.28) were larger than 0.05, which means that we cannot reject the null hypothesis (i.e. the H0 that the variables are exogenous is supported). Therefore, the DWH test suggests that endogeneity is not a concern in our dataset.
6. Discussion
6.1 Research findings
Despite the 26 possible combinations among the six causal conditions, only a few of the combinations were found to produce superior or inferior app performance. Meanwhile, the same causal condition (e.g. service incremental changes) and even the same pair of causal conditions (e.g. product radical changes and product incremental changes) can present in both a high-performance and low-performance combinations. These findings evidence the existence of complex interdependencies among the causal conditions. In this section, we infer effective or ineffective app update strategies by making sense of the complex interdependencies embedded in the analysis results. Following prior fsQCA result discussions (e.g. Campbell et al., 2016), we link the analysis results to both real-life examples in our sample and our theoretical underpinnings to develop intuitive and generalizable insights.
6.1.1 Effective mobile app update strategy
HP1: All three sub-combinations of HP1 include the presence of service radical change and the absence of app maturity as its core conditions. This shows the importance of service radical change to young apps' performance. The peripheral conditions indicate that service radical change can complement one of the other business element changes to produce superior performance for young apps, i.e. product radical changes in HP1a, product incremental changes in HP1b and service incremental changes in HP1c.
To gain a context-specific understanding about HP1, we revisited the sampled app strategic updates corresponding to the HP1 combinations. Characteristics of mobile technology surfaced as the key to the synergy between radical service change and the other business element changes in HP1-related app strategic updates. App strategic updates corresponding to HP1a and HP1b attempted to introduce much new product information via the small screen on mobile devices. As indicated by prior research (Ghose et al., 2012), consumers could experience difficulties in browsing, searching and comparing products on a small screen. A remedy to these technology-induced difficulties is technology-based new services such as location-based product recommendation, price comparison and social features (Pitt et al., 2011). A case in point from our sample is Ctrip. After the Ctrip app introduced four new product categories with numerous product selections in each category (e.g. hundreds of hotels in a newly added hotel deals category), it added new services to provide personalized product recommendations and location-based product ranking. These new services, coded as radical service changes in our analysis, facilitated consumers to narrow down their online experience with newly added product categories according to their offline contexts. With respect to HP1c, our sampled app strategic updates indicate that when a radical service change requires new knowledge from consumers, incremental service changes can serve as a bridge for consumers to grasp the new knowledge from their familiar knowledge. For example, when Ctrip pioneered the flight refund and reschedule functions in its app (service radical changes), consumers were concerned about the ease of use of these new services. The upgrade of voice recognition function and flight itinerary viewing function helped consumers draw on their familiar mobile web experience to utilize the new services.
Besides the synergistic combinations discussed above, the sub-combinations of HP1 show a consistent avoidance of simultaneous radical and incremental product changes. As suggested by HP1a and HP1b, introducing new product information via the small screen of mobile devices requires complementary new services. Simultaneous radical and incremental product changes can present a vast amount of new product information, which must come with a concerted bundle of new services to leverage mobile technology characteristics. However, a young app is unlikely to accumulate sufficient service features or app development experience to handle the complexity of coordinating service changes with simultaneous radical and incremental product changes. For example, in its early days, the Qunar app failed to achieve customer base growth when it simultaneously introduced overseas hotels and local tour products (product radical changes) and expanded ticket booking selections and domestic hotel selections (product incremental changes). Together, the insights regarding HP1a, HP1b and HP1c can be summarized as:
The baby-step strategy: Young apps can stimulate customer base growth by focusing on the complementarity between service radical changes and another business element changes such as product radical changes, product incremental changes, or service incremental changes and by avoiding a complex move including both radical and incremental product changes.
HP2: Apart from the difference on core conditions, HP2 also differs from HP1 by the presence of app maturity and rival app changes as two peripheral conditions. The combination of these two peripheral conditions indicates that HP2 is for mature apps in a highly competitive environment characterized by frequent updates of rival apps. Our sampled app strategic updates showed that service features of a mature app were often imitated by rival apps. Mature apps turned to product innovations, something not easily imitable by rivals, to attract new consumers. Mature apps can draw on their accumulated knowledge and experience to develop radical service changes that assist consumers to benefit from simultaneous radical and incremental product changes – a combination that must be avoided by young apps. For example, Fliggy – a mature app facing intense competition – expanded its tour package selections (product incremental changes) and introduced overseas hotel booking (radical product changes). Fliggy tapped into its app development knowledge to complement these product-related updates with brand new services such as real-time notifications during a tour and last-minute extensions of overseas hotel stay. This combination of app updates helped Fliggy attract more consumers than its rival apps.
HP2 requires the absence of incremental service changes to produce superior performance. This requirement can be attributed to HP2’s focus on introducing new products. In the sampled app strategic updates corresponding to HP2, radical service changes were clearly coupled with the consumption of a new category of products or new product selections in an existing product category (e.g. Fliggy’s real-time notifications for tour packages). Incremental services were not necessary for consumers to grasp the function of a radical service change. In summary, HP2 captures a mobile app update strategy:
The guarded-stride strategy: Mature apps in a highly competitive environment can stimulate customer base growth by combining product radical changes, product incremental changes and service radical changes, while avoiding service incremental changes.
HP3: Similar to HP2, HP3 includes the presence of focal app maturity. However, HP3 lacks rival apps' updates as a present condition, indicating its relevance to maturity apps in a low competition environment. A mature app had typically released a wide variety of updates and gained a stable customer base. To attract new consumers, the app must go above and beyond the strategic updates for a young app. The low competition environment reduces the risk for a mature app to experiment with a complex combination of app updates. Therefore, the mature app can take full advantage of its knowledge and experience to maximize synergistic relationships among all business element changes. For example, when Qunar became mature and faced few rival app strategic updates, its new app release simultaneously updated the app’s search function and order function (service incremental changes), introduced real-time notifications and discovery functions (service radical changes) and added new insurance products (product radical changes) and expanded ride-hailing options (product incremental changes). Such a constellation of updates enabled Qunar’s app to gain superior performance. The combination pattern of HP3 indicates:
The long-jump strategy: For mature apps in a low competition environment, a combination of product radical and incremental changes and service radical and incremental changes can stimulate customer base growth.
6.1.2 Ineffective mobile app update strategy
Compared with the high-performing combinations, low-performing combinations consistently exclude radical service change. Without the foundation of radical service changes, only a few business element changes could be orchestrated, which was not sufficient to produce ideal performance. In particular, LP1 indicates young apps' strategic updates that include only radical and incremental product changes. The product information is overwhelming or confusing for consumers without the assistance of complementary new services. For example, when Tongcheng strategically updated its app from version 5.2.0 to version 5.3.0, it introduced various overseas travel packages and cruise lines, and expanded flight selections and ticket booking selections but did not add new services. Such updates did not take advantage of mobile technology characteristics.
Our econometric analysis showed that LP1 is the only low-performance combination that has a statistically significant impact on app performance. The prominence of LP1 verifies the importance of specifying highly influential decisions early and correctly. LP1 also confirms the insufficiency of using only product innovation to achieve growth in the mobile platform (Ghose et al., 2012). With insights from LP1, we propose:
Ineffective strategy: For young apps in low competition, a combination of product radical changes as the influential condition, products incremental changes as the periphery condition and a lack of supporting service changes can hinder customer base growth.
6.2 Theoretical contribution
This study makes contributions to the literature in several ways. First, by embracing the interdependency among app strategic update decisions, our research bridges the grand theory of interdependent strategic decisions and phenomenon-specific observation of the interdependency of mobile technology characteristics. A general takeaway from the effective or ineffective strategies for mobile app updates is that a complex strategy problem in the mobile platform must be solved with respect to the constraints and capabilities of mobile technology. To date, research concerned with the complexity of strategy has not incorporated characteristics of mobile technology while mobile app research has played down the complexity of app strategic updates. Against this research background, our study identifies interdependency as a conceptual bridge between the strategy and mobile app research literature. We demonstrate the possibility to synthesize the grand theory of interdependent strategic decisions and phenomenon-specific observation of interdependency among mobile technology characteristics into a middle-range theory of mobile app update strategy. By answering to the what, when and how questions of mobile app update strategy, our proposed middle-range theory reveals mobile-app-specific solutions to the complex strategy problem.
Based on the middle-range theory of mobile app update strategy, the second contribution of our study is to develop an empirically testable version of the landscape search model. Previous landscape model studies mainly employ context-independent computer simulations as the research method to test the complex interdependency among strategic decisions (Ganco et al., 2020; Baumann and Siggelkow, 2013; Ghemawat and Levinthal, 2008; Levinthal, 1997). In contrast to these studies, we mapped the key concepts of a simulation model of the rugged performance landscape to measurable variables in a business context. Our study further identified fsQCA as an appropriate method to analyze combination patterns of the variables and reveal complex causality among them. This empirical approach paves the way for future landscape search models to tap into the increasingly available digital trails of business strategic moves. The abstract insights of landscape search can be contextualized, verified and extended into technology-based business phenomena. Therefore, our study not only demonstrates the generalizability of the landscape search model from computer simulations to empirical analysis but also provides feasible means for future empirical research.
We contribute to the mobile app research literature by moving beyond a linear view of the relationship between app update frequency and app performance. Prior mobile app research mainly focuses on the frequency of mobile app updates and emphasizes its linear effects on app performance (e.g. Comino et al., 2016; Gutt et al., 2019; Tian et al., 2020), but overlooks the complex interdependency of app strategic updating decisions and their nonlinear effects on app performance. Departing from prior studies, this study conceptualizes mobile app strategic updates as strategic moves in search for performance peaks in a rugged landscape, which opens the black-box of decision-making involved in app updates. Moreover, being different from the linear view in prior studies, the combination patterns of the strategic update decisions identified from our empirical study reveal synergistic or conflicting interdependency among app updates. Compared with update frequency in prior studies, the understanding of complex interdependencies inferred from our empirical findings are more likely to manifest how mobile technology characteristics shape strategy. Our landscape search model of mobile app update strategy, therefore, answers the recent call to theorize mobile-specific business strategy (Pitt et al., 2011).
6.3 Practical implication
For managers in charge of mobile app strategic updates, our study contributes a few actionable strategies to follow and an ineffective strategy to avoid. If their mobile apps are in the initial stage, managers should adopt the baby-set strategy to put the focus of update strategy on the complementarity between service radical changes and another business element change and need to avoid a complex move that includes both radical and incremental product changes. When mobile apps become mature, our study provides managers with two different types of update strategies. To resist the hyper-competition from rival apps' updates, managers could choose the guarded-stride strategy that combines product radical changes, product incremental changes and service radical changes, but excludes service incremental changes. In contrast, if their mobile apps are in a low competitive environment, managers should adopt the long-jump strategy to simultaneously combine product radical and incremental changes and service radical and incremental changes. Moreover, the ineffective strategy suggests that to avoid the failure update, managers should not exclude service changes if their young mobile apps radically and incrementally change products in a low competitive environment.
Besides the identified formulaic strategies, we provide another two general guidelines. First, the appropriate app update strategy depends on the maturity of the focal app and competitiveness of rival apps. The importance of these two factors reflects the wisdom that “if you know the enemy and know yourself, you need not fear the result of a hundred battles” (Sun, 2010). Therefore, mobile app developers should not only execute different update strategies at different stages, but also pay close attention to the strategic updates of competing mobile apps. Second, managers should look for synergy between mobile technology and business offerings. The characteristics of mobile technology such as small screen, limited storage and high portability provides managers with new opportunities for service innovation, including location-based product recommendation, price comparison, location-based product ranking. Theses business innovation that best take advantage of characteristics of mobile technology can be a universal foundation for all app strategic updates. Meanwhile, a technology-driven foundation has to be combined with enough other business elements to deliver novel consumption experiences deemed accessible by mobile consumers. Our findings suggest that managers should combine service innovation with other changes to realize the synergies of mobile apps' strategic updates.
7. Conclusion
As mobile apps become a new business competitive arena, companies must continuously update their mobile apps to attract new customers to their apps. The objective of this study is to identify the superior or inferior formals for mobile app updates. We employ the landscape search model from organization and strategy research to develop our research model. This study combines fsQCA with regression analysis to test the research model and identifies three effective and one ineffective mobile app update strategies. We linked the findings to strategic landscape search theory and developed generalizable app update strategies. A general takeaway from the effective or ineffective strategies for mobile app updates is that a complex strategy problem in the mobile platform must be solved with respect to the constraints and capabilities of mobile technology. Research findings highlight how characteristics of mobile technologies reshape business strategy. This study extends organization and strategy research to the mobile commerce contexts. We also demonstrate the generalizability of the landscape search model from computer simulations to empirical analysis.
Our study also has several limitations that warrant future research attention. First, our data were collected from a single market. This choice of research context ensures the rivalry relationship among the sampled apps. Future research can test the generalizability of our findings in other markets. Second, we employed the growth rate of app downloads as the performance outcome. The growth rate can capture the intended effect of an app strategic update in attracting new consumers. Further study can test performance outcome measures directed at existing customers' app use behaviors.
This research is funded by the National Natural Science Foundation of China (Nos: 72101241 and 72293572), Fundamental Research Funds for the Central Universities, China University of Geosciences (Wuhan) (No: G1323541816) and an Insight Grant from Social Sciences and Humanities Research Council Canada (No: 435-2017-0138).
Overview of concepts in our research model
| Strategic decision | Concept | Definition | Reference |
|---|---|---|---|
| What core business elements should be changed in app strategic updates? | Products | Tangible or intangible goods sold on a focal app | Cenfetelli et al. (2008) and Mocker et al. (2014) |
| Services | Supporting service functionality offered by a focal app | ||
| How substantial should core business elements be changed in an app strategic update? | Incremental change | Refinement of existing products or services with familiar knowledge | Fleming (2001) and Levinthal and Warglien (1999) |
| Radical change | Introduction of new product categories or services with new knowledge | ||
| When should an app strategic update be released in relation to the maturity of the focal app and its rival apps' updates? | Focal app’s maturity | The elapsed time after the initial launch of a focal app | Lee and Raghu (2014) and Tiwana (2015) |
| Rival apps' updates | The number of app strategic updates made by a focal app’s rival apps | Tanriverdi et al. (2010) and Tiwana (2015) |
Source(s): Authors' own creation/work
Descriptive statistics and correlations
| Variables | Mean | S.D. | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|---|---|
| 1. PRCs | 0.68 | 1.17 | ||||||
| 2. PICs | 0.58 | 1.00 | 0.13* | |||||
| 3. SRCs | 2.50 | 2.69 | 0.19** | 0.21** | ||||
| 4. SICs | 2.20 | 2.25 | 0.22** | 0.14* | 0.45** | |||
| 5. FoM | 45.71 | 12.51 | 0.05 | 0.08 | 0.20** | 0.33** | ||
| 6. RaUs | 39.96 | 18.88 | −0.02 | 0.05 | 0.30** | 0.32** | 0.53** | |
| 7. AP | 0.18 | 0.23 | 0.03 | 0.01 | 0.06 | −0.04 | −0.28** | −0.12 |
Note(s): **p < 0.01, *p < 0.05; PRCs=Product radical changes; PICs=Product incremental changes; SRCs=Service radical changes; SICs=Service incremental changes; FoM = Focal app’s maturity; RaUs = Rival apps' updates; AP=Focal app performance (the growing rate of focal app’s monthly downloads)
Source(s): Authors' own creation/work
High-performing combinations
| Combination | HP1 | HP2 | HP3 | ||
|---|---|---|---|---|---|
| HP1a | HP1b | HP1c | HP2 | HP3 | |
| Product radical changes | ⊗ | • | ⊗ | • | • |
| Product incremental changes | • | ⊗ | • | • | |
| Service radical changes | • | • | • | • | • |
| Service incremental changes | • | ⊗ | • | ||
| Focal app’s maturity | ⊗ | ⊗ | ⊗ | • | • |
| Rival apps' updates | • | ⊗ | |||
| Consistency | 0.84 | 0.85 | 0.82 | 0.84 | 0.89 |
| Raw coverage | 0.19 | 0.20 | 0.26 | 0.08 | 0.09 |
| Unique coverage | 0.02 | 0.07 | 0.07 | 0.02 | 0.02 |
| Overall solution consistency | 0.81 | ||||
| Overall solution coverage | 0.44 | ||||
Source(s): Authors' own creation/work
Low-performing combinations
| Combination | LP1 | LP2 | LP3 | |
|---|---|---|---|---|
| LP1 | LP2a | LP2b | LP3 | |
| Product radical changes | • | ⊗ | ⊗ | • |
| Product incremental changes | • | ⊗ | ⊗ | |
| Service radical changes | ⊗ | ⊗ | ⊗ | ⊗ |
| Service incremental changes | ⊗ | • | • | |
| Focal app’s maturity | ⊗ | • | • | • |
| Rival apps' updates | ⊗ | • | ||
| Consistency | 0.80 | 0.83 | 0.83 | 0.84 |
| Raw coverage | 0.07 | 0.27 | 0.13 | 0.17 |
| Unique coverage | 0.04 | 0.14 | 0.02 | 0.10 |
| Overall solution consistency | 0.84 | |||
| Overall solution coverage | 0.43 | |||
Source(s): Authors' own creation/work
Results of the first-stage selection model
| Isupdate | |
|---|---|
| Intercept | −1.63* (0.85) |
| Number of sibling apps | −0.15** (0.07) |
| Focal app’s maturity | 0.05** (0.02) |
| Rival apps' updates | 0.02** (0.007) |
| Investment relationship | 0.25 (0.37) |
| Download base of prior month | −0.000012* (6.30e−06) |
| Pseudo R2(%) | 10.82% |
| LR χ2 | 25.26*** |
Note(s): Standard errors are in parentheses. The dependent variable (Isupdate) is a dummy variable that indicates whether a company updates its app
Source(s): Authors' own creation/work
Results of the second-stage outcome model
| Combination effects model | Independent effects model | |||
|---|---|---|---|---|
| FGLS | OLS-PCSE | FGLS | OLS-PCSE | |
| Intercept | −9.54 (6.10) | −8.13 (6.38) | −10.48* (6.29) | −9.07 (6.64) |
| Investment relationship | 0.04 (0.03) | 0.02 (0.05) | 0.03 (0.03) | 0.03 (0.05) |
| Focal app’s updating version numbers | 0.00 (0.01) | −0.00 (0.01) | 0.01 (0.01) | 0.01 (0.01) |
| Promotion activity | 0.00 (0.02) | 0.02 (0.03) | 0.01 (0.01) | 0.03 (0.03) |
| Download base of prior month | −8.08e−07 (6.08e−07) | −9.22e−07 (7.88e−07) | −6.75e−07 (5.96e−07) | −1.26e−06 (8.78e−07) |
| Inverse mills ratio | 0.12 (0.10) | 0.13 (0.14) | 0.08 (0.09) | 0.05 (0.15) |
| Focal app’s maturity | −0.02* (0.01) | −0.02 (0.01) | −0.02** (0.01) | −0.02* (0.01) |
| Rival apps' updates | 0.00 (0.00) | 0.00 (0.00) | 0.00 (0.00) | 0.00 (0.00) |
| HP1 | 0.09** (0.04) | 0.10* (0.06) | ||
| HP2 | 0.25*** (0.06) | 0.23** (0.11) | ||
| HP3 | 0.26*** (0.08) | 0.49*** (0.14) | ||
| LP1 | −0.17* (0.09) | −0.28* (0.15) | ||
| LP2 | −0.00 (0.04) | −0.05 (0.06) | ||
| LP3 | −0.01 (0.04) | −0.02 (0.06) | ||
| Product radical changes | 0.00 (0.01) | −0.00 (0.01) | ||
| Product incremental changes | 0.00 (0.01) | 0.00 (0.01) | ||
| Service radical changes | 0.01* (0.00) | 0.01 (0.01) | ||
| Service incremental changes | −0.00 (0.00) | −0.00 (0.01) | ||
| R2(%) | 19.85% | 13.34% | ||
| Wald χ2 | 69.99*** | 59.51*** | 34.66** | 29.57** |
Note(s): (1) All variable correlations are below 0.6, except for the correlation between focal app’s maturity and download base of prior month. The highest score of variance inflation factors (VIFs) is 2.52, which indicates that multicollinearity is not a serious concern for our analyses (Chi et al., 2010). (2) Standard errors are given in parentheses. (3) *p < 0.1; **p < 0.05; ***p < 0.01
Source(s): Authors' own creation/work
Robustness tests for econometric analysis
| Combination effects model | Independent effects model | |||
|---|---|---|---|---|
| Fixed effect | Random effect | Fixed effect | Random effect | |
| Intercept | −1.31 (4.56) | −1.30 (4.49) | −2.55 (4.78) | −2.49 (4.71) |
| Investment relationship | 0.01 (0.05) | 0.01 (0.05) | 0.03 (0.05) | 0.03 (0.05) |
| Focal app’s updating version numbers | 0.02 (0.01) | 0.02 (0.01) | 0.03* (0.02) | 0.03* (0.02) |
| Promotion activity | 0.02 (0.03) | 0.02 (0.03) | 0.04 (0.03) | 0.04 (0.03) |
| Download base of prior month | −1.40e−06 (9.75e−07) | −1.40e−06 (9.75e−07) | −2.22e−06 (1.02e−06) | −2.22e−06 (1.02e−06) |
| Inverse mills ratio | 0.18 (0.15) | 0.18(0.15) | 0.11 (0.15) | 0.11 (0.15) |
| Focal app’s maturity | −0.00 (0.01) | −0.00(0.01) | −0.01 (0.01) | −0.01 (0.01) |
| Rival apps' updates | 0.00 (0.00) | 0.00(0.00) | 0.00 (0.00) | 0.00 (0.00) |
| HP1 | 0.16** (0.07) | 0.16** (0.06) | ||
| HP2 | 0.27* (0.14) | 0.27* (0.14) | ||
| HP3 | 0.46*** (0.16) | 0.46*** (0.16) | ||
| LP1 | −0.29** (0.15) | −0.29** (0.15) | ||
| LP2 | −0.03 (0.07) | −0.03 (0.07) | ||
| LP3 | −0.03 (0.07) | −0.03 (0.07) | ||
| Product radical changes | −0.00 (0.01) | −0.00 (0.01) | ||
| Product incremental changes | −0.00 (0.02) | −0.00 (0.02) | ||
| Service radical changes | 0.01 (0.01) | 0.01 (0.01) | ||
| Service incremental changes | −0.01 (0.01) | −0.00 (0.01) | ||
| R2(%) | 19.63% | 21.87% | 11.92% | 14.68% |
| Wald χ2 | 3.76*** | 64.68*** | 2.42*** | 40.10*** |
Note(s): Standard errors are given in parentheses; *p < 0.1; **p < 0.05; ***p < 0.01
Source(s): Authors' own creation/work
© Emerald Publishing Limited.