II. Previous literature and hypothesis

Sake has been analyzed from various perspectives, including its history (Kitagaki and Kitamoto, Reference Kitagaki and Kitamoto2013), corporate survival (Sasaki and Sone, Reference Sasaki and Sone2015), marketing (Lee and Shin, Reference Lee and Shin2015), tradition and new value-creation (Ishizuka et al., Reference Ishizuka, Kuo-Che and Kishi2022), apprenticeship (Hori et al., Reference Hori, Hoshino and Shimizu2020), and recent influence of wine culture (Tseng and Kishi, Reference Tseng and Kishi2023; Wang, Reference Wang2019) have been analyzed from various perspectives. However, there have been virtually no empirical studies so far analyzing the quality of sake brewing at the brewery level.

While sake brewing is a team effort, there has been no study about what kind of team produces high quality results. Team familiarity has been paid attention to because team performance does not fully depend on the team's past performances and task experience (Huckman et al., Reference Huckman, Staats and Upton2009). Team familiarity has been actively examined since the 2000s. However, its roots go back further. One of the key points is learning. When team members repeatedly work with one another, they learn with each other and develop a better way of collaborating (Goodman and Leyden, Reference Goodman and Leyden1991). This has been called social/organizational capital.

Empirical studies have observed positive effects of team familiarity on team performance in different fields, such as software development (Espinosa et al., Reference Espinosa, Slaughter, Kraut and Herbsleb2007; Huckman and Staats, Reference Huckman and Staats2011; Huckman et al., Reference Huckman, Staats and Upton2009; Staats, Reference Staats2012), basketball teams (Grijalva et al., Reference Grijalva, Maynes, Badura and Whiting2020; Sieweke and Zhao, Reference Sieweke and Zhao2015), and medical (Avgerinos and Gokpinar, Reference Avgerinos and Gokpinar2017; Maruthappu et al., Reference Maruthappu, Duclos, Zhou, Lipsitz, Wright, Orgill and Carty2016; Stucky and De Jong, Reference Stucky and De Jong2021). However, much of the previous literature has measured productivity, such as postdelivery defects, effort deviation, and team coordination error, rather than innovation. As team members share experiences, they build strong ties, positive social acceptance, and psychologically safe environments, which promote creative problem-solving (Edmondson, Reference Edmondson1999; Gruenfeld et al., Reference Gruenfeld, Mannix, Williams and Neale1996; Lee et al., Reference Lee, Edmondson, Thomke and Worline2004; Sosa, Reference Sosa2011).

However, a trade-off between novelty and productivity has been observed repeatedly in the literature on innovation, for example in the automobile (Abernathy, Reference Abernathy1978), hard disk drives (Christensen, Reference Christensen1993), shipbuilding (Greve, Reference Greve2007), and software (Ikuine, Reference Ikuine2022). If team members learn and coordinate efficient ways of working together as they repeatedly work together, it may promote efficiency but hamper new exploration (March, Reference March1991). Therefore, this study focuses on the effects of team familiarity on quality superiority rather than productivity.

Familiarity is defined as “the knowledge that members of a team have about the unique aspects of their work” (Goodman and Garber, Reference Goodman and Garber1988). Such knowledge encompasses not only the understanding of team members but also the task itself (Littlepage et al., Reference Littlepage, Robison and Reddington1997). As each task requires unique alignments of machinery, environmental conditions, and operational activities, task familiarity is considered to positively influence task performance across various domains, including mining (Goodman and Leyden, Reference Goodman and Leyden1991), software development (Banker and Slaughter, Reference Banker and Slaughter2000; Walz et al., Reference Walz, Elam and Curtis1993), and medical sector (Reagans et al., Reference Reagans, Argote and Brooks2005). Besides, task familiarity represents a pivotal resource for new product development performance (Wu et al., Reference Wu, Rivas and Chen2019). As mentioned above, task familiarity is specifically related to the knowledge about team members' principal work (Harrison et al., Reference Harrison, Mohammed, Mcgrath, Florey and Vanderstoep2003). The more members in a team with high task familiarity, the faster others will acquire knowledge related to any task. This, in turn, accelerates knowledge exchange within the entire team (Wu et al., Reference Wu, Rivas and Chen2019).

While undoubtedly essential to achieving organizational goals, task familiarity can also have detrimental effects on team performance in novelty exploration. Arrow and Mcgrath (Reference Arrow and Mcgrath1993) explore the adverse consequences of heightened task familiarity on team performance, thereby elucidating potential challenges and constraints in the realm of team dynamics (Arrow and Mcgrath, Reference Arrow and Mcgrath1993). Task familiarity can precipitate complacency within a team (Choi et al., Reference Choi, Delise, Lee and Neely2021). When team members have extensive experience in their assigned roles and tasks, their motivation to explore novel approaches or innovations may diminish. This complacency can stifle creativity and limit problem-solving capabilities, as individuals may gravitate toward familiar methods instead of considering innovative solutions (Choi et al., Reference Choi, Delise, Lee and Neely2021; Goodman et al., Reference Goodman, Devadas, Hughson, Campbell and Campbell1988; Wood and Lynch, Reference Wood and Lynch2002). Moreover, elevated task familiarity can result in a lack of cognitive diversity within a team (Harrison et al., Reference Harrison, Mohammed, Mcgrath, Florey and Vanderstoep2003; Mohammed and Nadkarni, Reference Mohammed and Nadkarni2014). In addition, in a team characterized by excessive task familiarity, there is a tangible risk of stagnation in learning and skill development. The absence of exposure to new challenges and opportunities may hinder team members from acquiring new competencies and knowledge, thereby imperiling the team's overall adaptability and capacity for growth (Eddy et al., Reference Eddy, Sydow, Alfred and Garza-Mitchell2015). Based on this, this paper proposes the following hypothesis, considering that task familiarity would have a negative impact on novelty-creating team performance.

III. Sake brewing team and quality competition

Sake is an alcoholic beverage made from rice, rice malt, yeast starter, and water, which are fermented and then strained. Sake brewing is highly knowledge-intensive. One of the steps requiring a high level of expertise is the multiple parallel fermentation processes. These simultaneous multiple fermentation processes are defined as the two processes of saccharification and fermentation: enzymes in the rice malt turn starch into glucose, and glucose is fermented into alcohol using yeast, proceeding in parallel within manufacturing-use containers. For the manufacture of sake, it is necessary to maintain a good balance between saccharification and fermentation while proceeding with the fermenting process.

A. Sake brewing team

Sake brewing is conducted by a team of brewers supervised by a master brewer called Toji. The master gets a contract to brew sake from the sake brewery via a master brewer association's brokerage and takes full responsibility for sake brewing. The master brewer selects and hires his/her team members. Since sake brewing is seasonal, beginning in late September or early October and ending in March, the contracts between the brewery and the master brewer, and between the master brewer and his team member, are on an annual basis. The master brewer assigns his/her chosen team members to roles associated with the sake brewing process. The breweries work together to brew sake at the sake brewery.

Including the master brewer, brewers usually started with simple tasks and gradually engaged in skillful tasks over several years (Hori et al., Reference Hori, Hoshino and Shimizu2020). Through brewing supervised by the master brewer, chosen brewers not only develop task familiarity but also gain an understanding of the brewing process and team management. The brewing tasks are divided in a hieratical manner. As the brewers gain skills, they advance the rank hierarchical career ladder. A typical hieratical ranking system corresponding to each task in the sake brewing process is illustrated in Figure 1. The number on the left of the title signifies the career ladder of brewers. Brewers start their careers as an assistant and move up the ladder. While the number of team members varies depending on the size of the brewing, the typical team size is around five brewers, which corresponds to the sake brewing process illustrated in Figure 1.

Figure 1.

Job title, task, and process.

As described above, the master brewer is a director of brewing who takes full responsibility for brewing at the brewery. An assistant master brewer plays a supportive role to the master brewer. A polisher shaves off the rice converted from brown to white. A steamer washes and steams the rice. These processes are one of the most important processes in raw material processing. The amount of steam is adjusted according to the condition of the rice. A rice steamer checks the steaming conditions together with the master brewer. Then, a malt maker is in charge of making molt (the rice in which Aspergillus oryzae is grown). This manufacturing process requires a relatively high level of expertise, as the result of this process has a significant effect on the quality of the sake. Alongside malt making, a fermentation starter producer is responsible for making a fermentation starter or yeast starter. When adding ingredients, a person in charge of the fermentation process takes control of temperature regulation and facilitates fermentation to make unrefined sake. A presser squeezes the fermented unrefined sake. Newly joined workers perform odd jobs such as cleanup and assistance in the processes noted above. Under this division of labor, sake is brewed as a new product every year.

IV. Data and estimate strategy

Two datasets are needed to examine the impact of team familiarity on innovation. The first dataset is to measure team familiarity. To measure team familiarity, we need the number of times a member has worked with another member in the past. In other words, we need data on the transition of team members over time.

The annual brewer lists of the Nanbu Master Brewer Association (Nanbu Toji Kyokai) allow identifying team members and their role. The association, whose precursor was established in 1914, is the largest master brewer association in Japan. The Nanbu Toji Association accounts for nearly 30% of all master brewers in Japan. There are approximately 30 brewers' labor associations in Japan. These associations are organized geographically. The brewer list published annually provides detailed information on all registered brewers, such as their home address, their birthday, the branch to which they belong, tasks they perform, the brewery they work for, and its location. The list allows identifying the team members, the role of members, and their career path. Based on the brewer lists, this study measures team familiarity and task familiarity.

The second dataset is to measure the performance of the team, innovation in particular. This study uses the results of the Japan Sake Awards from 1956 to 2018. The advantage of using this data is that, since the awards are given in a blind tasting, the team's performance can be viewed without the influence of the master brewer on the team or the brewery's reputation. Furthermore, only one sake from a single brewery can be entered for this award, so the team and its performance can be accurately matched.

With these two datasets, this study estimates how team familiarity is associated with innovation by the following model:

\begin{align*}Awar{d_{it}} =& {\beta _0} + {\beta _1}Team{\text{ }}Familiarit{y_{it}} + {\beta _2}Task{\text{ }}Familiarit{y_{it}} + {\beta _3}Team{\text{ }}Siz{e_{it}} \\ & + {\beta _4}Brewery{\text{ }}Awar{d_{it}} + {\beta _5}Brewer{\text{ }}Awar{d_{it}} + {\lambda _i} + {u_{it}}.\end{align*}

The dependent variable, $Awar{d_{ijt}}$, is the dummy variable which takes 1 if a brewing team i in year t wins the gold medal at the national quality competition. Otherwise, it takes 0.

The primary independent variable of interest is Team Familiarity. Following the previous literature (Huckman et al., Reference Huckman, Staats and Upton2009), this study measures team familiarity with the following steps. First, we calculate the number of times each pairing of team members i and j have worked together before the current brewing. We sum this value $P{W_{ij}}$ over every unique pair on a team to capture team-specific experience, $\mathop \sum \limits_{i = 1}^N \mathop \sum \limits_{j = 1}^N P{W_{ij}}/N\left( {N - 1} \right)2$, where N is team size. Working with the exact same team member for an additional year raises the Team Familiarity by 1.

We introduce two variables to test how the task familiarity of team members influences team performance. The first is a variable to measure the experience of the master brewer. The variable is Brewer Task Familiarity, which counts how many years have passed since a master brewer served as a brewer for the first time. We measure it by using the list of master brewers provided by the Nanbu Master Brewer Association. The second variable, Individual Task Familiarity, measures the average task familiarity of team members. As described above, sake brewing is carried out through the division of labor. Each team member has a different experience. For example, some members may have been assistants or steamers for a long time, while others may have been promoted to master brewer with relatively few years of experience in their respective tasks. Therefore, we measure the task familiarity of team members in five distinctive brewing process categories illustrated in Figure 1. The five categories are represented by the subscripts of the variables. We construct the task familiarity variables as follows. Individual Task Familiarity 5 is the average number of years team members experienced as master brewers. Individual Task Familiarity 4 is the average number of years team members experienced as assistant master brewers or rice malt makers. Individual Task Familiarity 3 is the average number of years team members served as fermentation starter makers or persons in charge of the fermentation process. Individual Task Familiarity 2 is the average number of years team members served as a steamer, polisher, or presser. Individual Task Familiarity 1 is the average number of years team members served as any other assistant.

In this paper, we control for individual and organizational capabilities that affect novelty-generating team performance by introducing the following two variables. By using the dataset, Brewery Award is a variable measuring how many times a brewery previously won the award. We control the master brewer's capability by introducing Brewer Award, a variable measuring how many times a master previously won the award by using the Annual Japan Sake Awards' results.

Since many of the breweries are small businesses and unlisted firms, they do not disclose financial information such as sales, operating income, and capital. Therefore, it is difficult to directly measure the size of the brewery. However, we introduce Team Size, which counts the number of brewers of a brewing team i in year t. Since geographical conditions such as industrial clustering, climate, and water quality can affect the quality of sake brewing, we control the location of the sake brewery at the prefecture-level as λ.

This paper analyzes the impact of familiarity of the brewing team on brewing performance. In other words, we assume that the cause and effect is the composition of the team members and the result is the performance of the brewery. However, the composition of team members is not randomly determined. A brewery that wants to improve its sake brewing performance may change the composition of its team to achieve that goal. In such cases, it is not only the composition of the team that affects sake brewing performance, but also the possibility of reverse causality. For this reason, we use an external shock, the Great East Japan Earthquake of 2011, as a control variable to estimate causality. The earthquake caused much damage mainly in the Tohoku region. We assigned a dummy value of 1 to the brewers who lived in the special evacuation zone, where the earthquake caused significant damage and the residents were asked to evacuate, and a dummy value of 0 to the other brewers. As those who lived in the special evacuation zone had to move, it is thought that many of them were unable to continue making sake in the same brewery as before. Therefore, it can be assumed that the team familiarity of breweries that employed brewers from the affected areas would be reduced because it would be difficult for the breweries to maintain their original human resources, while the team familiarity of breweries that did not employ brewers from the affected area would not be changed by the earthquake. Since the effects of the earthquake are exogenous to the breweries and affect team composition, but are not expected to directly affect team performance, they are used as an instrumental variable to estimate two-stage probit models to mitigate endogeneity. We use this exogenous shock to examine how team familiarity affects team performance. In the first step, we run an Ordinary least squares (OLS) regression using a dummy variable, Areas Damaged by Earthqake, for the special evacuation zone against the endogenous variables, which are Team Familiarity and Task Familiarity, and obtain the predicted values of the endogenous variable. In the second stage, we incorporate the predicted values from the first stage into the probit model to estimate the relationship between the dependent variable and the explanatory variables.

V. Empirical results

With the Annual Japan Sake Awards and the Nanbu Master Brewer Association's master brewer list, we can cover the quality competition results from 1956 to 2018 and identify 11,866 brewers and 1,003 breweries. Table 1 shows the descriptive statistics of the variables and correlation matrix.

Table 1.

Descriptive statistics and correlation matrix

The mean value of Team Familiarity is 2.582, and its standard deviation is 2.798. The average year of team members' experience in brewing, Brewer Task Familiarity is 12.713 years, and its standard deviation is 11.566 years. The brewer with the most years of experience has 65 years of experience. It is interesting to see the task familiarity when the task familiarity of the brewers is broken down into five tasks. As one moves up from task 1–5, the average year of experience also increases. As noted above, as one moves up from tasks 1–5, the tasks become progressively more advanced. In other words, more advanced tasks require a longer period to become skilled. The correlation coefficient table does not reveal many particularly high correlations among the variables.

Table 2 shows the estimation results of the analysis. We use the logit model to estimate how team familiarity and task familiarity influence team performance because the dependent variable is the dummy variable which takes 1 if a brewing team i in year t wins the gold medal at the national quality competition; otherwise, it takes 0. The results report the odds ratio rather than the coefficient.

Table 2.

Estimation results, logit model

Clustered standard error in parentheses,

*** p < 0.01, **p < 0.05, *p < 0.1

Models 1 and 2 are baseline models for team familiarity; Model 2 controls the location of the brewery at the prefecture level, which is not done in Model 1. As we will see later, the odds ratio for Team Familiarity is below 1 in all models. This result supports the hypothesis H1b.

Models 3 and 4 introduce task familiarity: Model 3 examines Brewer Task Familiarity, which is a proxy for the familiarity on the more general brewing tasks, and Model examines Individual Task Familiarity 5–1, which is a proxy for the familiarity on the more detailed subtasks. The odds ratios of task familiarities, both general and broken down into subtasks, are below 1, which means a negative association with award winning. This result supports hypothesis H2b.

From Model 1–7, all models but Model 4 shows the negative association between team familiarity and award winning with statistically significance. The results of task familiarities are negative as well. Since some studies have indicated the inverted U-shape relationship between team familiarity and team performance (Katz, Reference Katz1982; Sieweke and Zhao, Reference Sieweke and Zhao2015), we introduce a squared term for team familiarity and task familiarity in each and examine the possibility that the relationship is U-shaped in Models 5, 6, and 7. The results suggest a U-shaped relationship, since the squared terms for Team Familiarity and Task Familiarity are statistically significant and have odds ratios greater than 1. The odds ratios of the variable indicate that such a U-shape relation exists. However, they show that the upward effect as suggested by U-shaped relationships is marginal.

Looking at the controlled variables, the odds ratios for Team Size, Brewery Award, and Brewer Award are all exceeded 1, indicating a positive relation. The odds ratio is particularly great for Brewery Award and Brewer Award, indicating that excellent breweries and team members who have achieved outstanding results in the past have a high probability of winning awards the next time around, even though the competition is based on blind tasting by experts so that reputation effects can be excluded.

Next, we perform robustness checks using four different models with two-stage probit models. In Table 3, we present the results of the first and second stage estimations for Models 8 to 11, respectively. The first stage estimations are OLS and report coefficients, while the second stage estimations use probit model and reports exponentiated coefficients, which are relative risk ratios (RRR). When RRR is greater than 1, it indicates that the probability of the dependent variable occurring increases as the independent variable increases, and when it is less than 1, it indicates that it decreases.

Table 3.

Robustness check, two-stage probit model

Clustered standard error in parentheses,

*** p < 0.01, **p < 0.05, *p < 0.1

In the first stage estimation of Model 8, there is a statistically significant negative association between Team Familiarity and Areas Damaged by Earthquake. And in the second stage results using the predictions from the first stage estimation, the RRR for Team Familiarity is 0.79, the same negative relationship observed here. In Model 9, the first-stage results show a negative association between Brewer Task Familiarity and earthquakes, but not statistically significant even at the 10% level. This result indicates that although earthquakes work well to address the endogeneity of team familiarity, caution should be exercised with regard to task familiarity. The results of the first stage for Models 9 and 11 show that there is no statistically significant relationship between earthquakes and task familiarity. The reason for this should be related to the brewery's choice of hiring new personnel. If the people who had been working are unable to come to work at the sake brewery due to the earthquake, the sake brewery will have to hire new personnel. Eventually, the value of team familiarity will be decreased, as this is confirmed in Models 8 and 10. However, if the earthquake prevents someone who has been working on a particular task from coming to work, the sake brewery will hire someone who is familiar with the same task. If this is the case, team familiarity will decrease, but task familiarity will remain the same.

Models 10 and 11 use the same estimation as Models 8 and 9, narrowing the sample to 1980 and later. The shorter period used in the analysis is due to the starting point constraints of our data. The brewer lists we use to calculate Team Familiarity were compiled from 1955. Therefore, 1955 is the first year for calculating Team Familiarity. In other words, if a team was brewing sake in 1955, we regard that as the first year the team was organized. However, it is possible that the team was organized before 1955. Figure 2 shows the time series of the Team Familiarity box plot from 1955 to 2018.

Figure 2.

Box plot of team familiarity time series.

It shows that the lower quantile, median, and upper quantile have been relatively stable since the 1980s. This is probably an effect of the fact that the team that had started brewing before 1955 was virtually no longer in existence at this time. The results for Model 10 with a restricted sample are generally consistent with those of Models 8. Areas Damaged by Earthquake in Model 11 does not function as an instrumental variable for Task Familiarity as in Model 9.