Tiny Use Case 2: Can we test one of the points from Hiroki Azuma’s “Otaku: Japan’s Database Animals” with the JVMG database? Part 5: Summary and lessons learned

Here we are, finally done with our series on Tiny Use Case 2, so let’s do a quick TL;DR summary of what happened in the four preceding blogposts (1, 2, 3, 4):

• • In part one we wanted to find a claim to test from Hiroki Azuma’s seminal book Otaku: Japan’s database animals (2009, Japanese original published in 2001) with the help of the JVMG database
  • We settled on the point captured in “As a result, many of the otaku characters created in recent years [the late nineties to two thousand] are connected to many characters across individual works, rather than emerging from a single author or a work.” (p 49)
  • And formulated two hypotheses based on this idea of characters becoming increasingly derivative of each other, later finalized in the following form:
    1. The portion of new characters with shared traits should increase over time.
    2. The portion of shared traits among new characters should increase over time.
  • Then in part two we introduced some of the specificities of the two datasets (The Visual Novel Database (VNDB) and Anime Characters Database (ACDB)) we were going to use for our analysis, and explained the operationalization of our concepts on the data.
  • For better comparability we decided on splitting the ACDB data into two datasets, one for visual novel characters and another for (mostly anime) other characters.
  • We identified a temporal trend in all three datasets with number of characters and average number of traits both peaking around the early 2010s and then declining. Assuming that the number of traits used to describe characters is a good proxy value for the amount of attention being paid to the data, we concluded that our datasets are most likely not complete, especially for the second half of the 2010s.
  • In part three we reported on the series of regression analyses we conducted to identify the best models explaining the change in the number of characters with shared traits in each of the three datasets, and found that our first hypothesis was not substantiated by our results. Furthermore, we found no adequate regression model for testing our second hypothesis.
  • Finally, in part four we first examined the validity of our results.
  • Then, we examined the theoretical implications of our findings on Azuma’s argument. We concluded that our results suggest considering the possibility that the production side of Japanese anime and manga has always been “database-like” in its mode of operation. This, however, would not invalidate Azuma’s main line of argument in relation to consumption and postmodernity in Japan. Nevertheless, amending his line of reasoning in relation to the production side of the argument would help foreground his potential theoretical debt to the work of Toshio Okada.

We saw in an earlier post that the third phase of each Tiny Use Case revolves around evaluation. Thus, it is time for us to evaluate what we have learned during the course of TUC 2, and what remains open for further exploration.

First of all, we have developed a better understanding of our databases, and have demonstrated that it is viable and fruitful to work on two different databases to increase the validity of our analyses. Second, it is clear that beyond our work on the accuracy of the data, we should also explore data quality in relation to its completeness. Third, we demonstrated that it is possible to use the available data to test research questions despite the lack of data completeness.

As for the remaining questions, of which there are many, we have already mentioned that further analysis of our second hypothesis in relation to which we found no viable regression models could definitely be revisited. We have also already pointed out that cleaning the VNDB data in relation to hierarchically nested traits might also help our analysis with reaching more valid conclusions. Furthermore, a more full-blown exploration of this case study – one that goes beyond the limits of a tiny use case – should probably entail a testing of the robustness of our results over a wider range of possible values that were fixed in our analysis. For example, would our results change, and if yes, how, if we were to set the number of shared traits required for us to consider two characters to be characters with shared traits to values other than five. Similarly we could conduct sensitivity analysis with different values of average number of shared traits, for the part of our analysis where we fixed that value at seven and eight for the two databases respectively.

Thus, as always, many possibilities remain to be explored. This is even more so the case for tiny use cases, which have a limited time span in which to be realized, in order to serve their purpose of providing feedback in the exploration, innovation and development cycle of the overarching JVMG project.