1. A single woman (26). She feels awkward in groups of women who are always and forever chatting. She would like to leave but knows that the relationships formed in such groups can be important for her future.
  2. A self-employed barber, male (43), married with three sons. Digital technology makes him uncomfortable. His car has a digital navigation system, but he always has someone else look at it and tell him where to drive. Else, he prefers a paper map.
  3. A married woman (43) with a husband and three sons would like to keep her house spic-and-span. It isn’t just that she doesn’t have the time or money, she never feels like doing the cleaning.
  4. A man (41), who lives in an apartment with his wife, a college classmate, and their fith-grade son.  Feels like a big third-grader who has never grown up. He has gotten older and fatter but doesn’t do what grownup men do, go to hostess bars, play golf, or gamble.
  5. A single man (33). Finished graduate school and works as an IT consultant. Tried living by himself for a year but has moved back in with his mother and sister.  He would like to be the cool guy who walks into a bar by himself and orders a drink,  but he is too shy to strike up a conversation with strangers.
  6. A woman (38), just got married last November (2011) after energetically pursuing a mate. She is wondering if she doesn’t dress too young for her age. Should she stop wearing short skirts? She still likes to dress like a single woman in her early 30s.
  7. A man (63), retired, lives with his full-time housewife spouse. He feels like zero without his company and the co-workers with whom he used to pal around. He has no community in the neighborhood.
  8. A single man (25). LIves with his parents, two sisters, and his grandmother in a six-person household. Finds it hard to converse with the people close to him but has no problem speaking up at work-related meetings organized by others.
  9. A woman (17) in her third year of high school. Weak at math and wonders why she has to study it.
  10. A man (50), a company employee who lives with his wife and two daughters. Feels like he doesn’t have the courage to stand up for what is right.
  11. A woman (70). She ran her own beauty shop for more than 40 years and now does volunteer work, cutting the hair of elderly people in nursing care. But she was brought up with a clear set of does and don’ts.  She loves sushi but is still too embarrassed to go into a sushi shop by herself.
  12. A man (17), in his third year of high school. Sings in the chorus and belongs to the calligraphy club. He says that it is hard to explain but that he is very uncomfortable calling himself “ore” (a rough, masculine way of saying “me”). He never uses this term for himself.

As I read the comments from these twelve individuals, I am struck by how ordinary they seem to me. Have I lived in Japan for too long? Or am I right to think that we can find people everywhere who feel that they have these kinds of weaknesses?

This project has been brewing for three years. The girls who became idols were usually straight, orthodox types. That left Yankees an untouched market. While preparing for the auditions, we conducted market research among Yankee girls living in the northern Kanto. We found that in Japan that there are still those that want to be “bad.”  Recently we have seen the emergence of the oraora “bad-bad” style, a blend of gyaru  “gal” (heavily made up and outrageously costumed) and the original Yankee style. In it we saw a foreshadowing of a full blown revival of the Yankee style that flourished in the 1980s.

To Aoki, the ideal image of a Yankee idol includes a military uniform or jumpsuit and a lavishly decorated car, in a blend of classic Yankee and modern idol taste. Seen from abroad as a kind of “cosplay” (costume play), the Yankee idol would be perceived as cool. The new girl group’s fans would include boys with the same oraora tastes, otaku deeply immersed in Akihabara culture, and “light otaku,” young men who may not at first glance appear to be otaku. Otaku and Yankee might appear to be opposite types [one passive, the other aggressive; one middle class, the other lower class], but they have much in common says Aoki.

The Avex Group aren’t the only ones interested in Yankees. The HakuhodoDY group has conducted its own research, using a sample of 800 men and women who are either former Yankees or former Gals, now aged 20-39. The sample was chosen from those who, when shown illustrations of typical figures representing the two subcultures, selected them as “close to what they themselves had looked like in high school.” The survey covered attitudes toward friends and family, information sources and media contact, and consumption trends. Results included closer than average attachment to people in the neighborhoods where they grew up and live and longer than average time spent watching television, but far less use of social networking sites (SNS). More than 60% say that they spend money when they have it and, while they have to scrimp and save, will spend on drinking and other treats.

Of particular interest is the emerging category of “Gal mamas,” who are now catered to by their own magazines. Marrying younger and having children sooner than other Japanese women, these former Gals are an attractive target for baby care, toy and other industries who have seen markets slump due to Japan’s low birth rate.

The special section includes a discussion by Koji Nanba (KN) and Satoshi Hamano (SH). Nanba is a sociologist and university professor who studies youth subcultures and the author of Yankii shinkaron (Yankee Evolution, Kobunsha, 2009). Hamano is a columnist for the Asahi Shimbun newspaper and the author of Aakitekuchanoseitaikei (The Ecology of Architecture, NTT, 2008) and AKB48 hakunetsuronso (The White Hot Debate about AKB48, Kobunsha, 2011). The following passages are selected and adapted from this discussion.

Yankees make up 20% of fans at AKB48 meet and greet events

Editor: Marketers tell us that we should pay attention to Yankee, but they find it difficult to create a convincing image of them as consumers.

KN: Most people who work in advertising or media are from white collar backgrounds. Without direct experience, they have a hard time coming up with a realistic image.

SH: You are talking about people like me who have never experienced the Yankee subculture.

KN: The music and dance group EXILE has been a major factor in the renewed interest in Yankee. There aren’t many people who call themselves Yankee. But there can be no question about it that Yankee taste has become a major phenomenon.

SH:  Yankee taste is part of the background to the success of AKB48 as well as EXILE. At events where fans get to meet AKB48 members, 20% of those who show up are Yankees. They come with the names of their favorite members of the group stitched to their uniforms.

KN: They seem to be imitating the bodyguard/ardent fan style from the 1980s. We thought the era of Yankee chasing after their favorite celebrities was over, but the new idol boom appears to have stimulated a Yankee resurgence.

SH: There is also the Majisukagakuen series on TBS, in which AKB48 members appear as Yankee characters, who fight, hit and scream at each other. (The third season is currently being broadcast from July to October.)

SH: Of those who show up at AKB48 events, only 10-20% are classic idol-otaku (men with an obsessive interest in their idols). Another 20% are women in their teens or early 20s. The overwhelming majority are high school or college boys, including those who play at being bad boys.

Editors: In the HakuhodoDY survey, AKB is the seventh most popular group among former Yankees and former Gals. They rank only number nine among ordinary subjects in the same age band.

KN: AKB48 may be why we see Yankee becoming involved with idols again.

White collar stagnation casts spotlight on determination to survive.

Editor: Yankee and otaku are sometimes seen as polar opposites. How has each type changed?

KN: In most cases the Yankee characters depicted in recent manga and movies have serious, practical issues to confront. In the video series,“Saitama Rapper,” things don’t turn out too badly in first two episodes, but the third and last episode is grim.

SH: In the manga series Yamigane Washijima-kun (Yamigane can be literally translated “Dark Money) has been made into a movie that typifies this “dark Yankee” trend. In it AKB48 member Oshima Yuko appears as the freeter (casual worker) heroine.

SH: To me these films reflect the effects of the recession and the recognition that hoowever hard the work, you have to do it to survive. When Japan seemed to be a “100 million middle class” country, everyone wanted to escape from blue collar lifestyles. White collar stagnation has made Yankee, the most visible blue collar subculture, attractive again.

KN: From a university professor’s perspective,you can’t help feeling the youthful strength and energy it he way that Yankee characters are depicted. Nowadays, university graduates without specialized skills, especially those with degrees in the humanities, get assigned to sales in the companies at hire them. But those with Yankee tendencies are likely to be better salespeople.

SH: The more I read your book Yankee Evolution, the more I am persuaded of the similarities between Yankee and otaku. Both have adopted bits of foreign culture but transformed them in a Japanese Galapagos way. Most Japanese manga and anime have their roots in SF or Disney.

KN: Yankee fashion is a mixture of the styles worn by off-duty American GIs in the 1960s and 70s with Japanese outlaw culture.

Desire to live real life to the full and school castes set the stage

Editor: How is Yankee consumption behavior changing?

SH: The earlier versions were fundamentalists or serious avant-garde. The ones we have seen since 200 are “otaku lite.” The AKB48 fans belong to this category.

KN: We no longer see those who are absolutely committed to their Yankee identity. I remember from a previous interview your using the expression “tagging”  to describe what has happened.

SH: It used to be that Yankee embodied a strong lower class consciousness. Otaku were also clearly defined as the unpopular ones by school caste systems. Now otaku lite are coming out of their shells and looking for something fulfilling in real life. As media coverage of Yankee items has risen,  it has become increasingly possible for otaku to think that simply purchasing certain items is the key to acceptance as one of the gang.

Consumption for the sake of relationship. Even otaku can belong to the gang.

SH: AKB48 illustrates how otaku turn out for concerts and chances to meet their idols. Genbashugi (an ideological committment to “being there”) makes participation in events and buying memorabilia part of who they are.

KN: The otaku lite are really different from the classic otaku who stays home watching anime all the time.

SH: They want to have a relationship, to be part of a group. Dentsu is now promoting its SIPS (Sympathize, Identify, Participate, Share and Spread) model for understanding social media. At the core of the SIPS process are relationships created by “chasing what is interesting” or “being bad.” One good example an Ezaki Glico confectionary campaign in which parts of faces from different AKB48 members are mixed to create a portrait of the ideal AKB48 girl.

KN: Whether Yankee or otaku, fans of the same idols have something in common that creates a relationship and makes communication possible. Adding a Yankee flavor to AKB48 is sure to boost their popularity.

Reading this discussion, I find myself thinking about kachigumi (winners) versus makigumi (losers), a classification that reflects the growing class polarization of Japanese society. Taking pride in being a bad ass is not a bad way to maintain self-esteem if opportunities to get ahead and escape from a life of repetitive office or physical labor are increasingly scarce. And it’s not, of course, an option restricted to young Japanese.

During the more than ten thousand years of the Jomon period, these clay figurines were produced throughout Japan. For this exhibition, examples from Hokkaido to Kyushu have been brought together here in Shigaraki. In their powerful presence, you can see the roots of the Japanese people.

  Here the network in question is from 2006, a year in which 2,662 creators participated in creating 808 winning ads. The colors distinguish Ads (Yellow) from Creators (Green). The sizes of the nodes indicate the number of edges connecting each node to its immediate neighbors. The colors of the lines indicate the roles that connect individual creators to the ads on which they worked. The result is unintelligible to the human eye.

  The inability of the human eye to parse large and richly coded networks means that, for practical purposes, most empirical network analysis becomes either computational, using software to crunch numbers and pondering the results, or requires dissection of large networks into subnetworks. In practice, these two approaches are frequently combined, and Pajek provides numerous tools for these purposes. Using Pajek, we can

find clusters (components, neighbourhoods of ?eimportant’ vertices, cores, etc.) in a network, extract vertices that belong to the same clusters and show them separately, possibly with the parts of the context (detailed local view), shrink vertices in clusters and show relations among clusters (global view).(Bataglej, 2008:8)

  At the same time, we can use the Info command to examine the numbers generated by Pajek during these and other operations. Continuing, then, with Ads-Creators 2006, the network illustrated in Figure 7, we begin with Info>Network>General.  In the Info report window we see the following table.

Number of vertices (n): 3469
——————————————————————————-
                      Arcs       Edges
——————————————————————————-
Number of lines with value=1       0       1079
Number of lines with value#1       0       6195
——————————————————————————-
Total number of lines           0       7274
——————————————————————————-
Number of loops               0         0
Number of multiple lines         0       1102
——————————————————————————-

Density1 [loops allowed] = 0.0012089
Density2 [no loops allowed] = 0.0012093
Average Degree = 4.1937158

2-Mode Network: Rows=808, Cols=2661
          Density [2-Mode] = 0.0033831

Reading from the top, we see that the network contains a total of 3469 nodes (here called vertices). If this were a directed network, we would have to distinguish between Arcs and Edges. In this network, however, there are no Arcs. All of the lines are undirected Edges and represent relations in which the relationship between node A and node B is the same as that between node B and node A.

  In the next block we see that the number of lines with value equal to 1 is 1079, while the number of lines with value not equal to 1 is 6195. Here, however, we have to be careful. The ways in which Pajek presents numbers like these are very general and highly abstract. Pajek only reports the results of calculations without differentiating what these numbers might mean. In some cases, the numbers of lines equal or not equal to one are structural properties resulting from calculations. Here, however,  the numbers are assigned codes that indicate the roles that connect Creators to Ads (1=Copywriter, 2=Creative Director, 3=Art Director, 4=Designer, 5=Photographer, 6=Planner, 7=Producer, 8=Film Director, 9=Cameraman,  and 99=Other). The numbers are, in other words, only labels in classification; they correspond to the line colors that appear in the network diagram and can be used to select particular sets of lines. They should not, however, be used in calculations.

  The next two blocks tell us that the total number of lines is 7,274, of which 1,102 are multiple lines. These numbers are significant, since they tell us that some pairs of nodes are connected more than once (in this case because the same creator may play multiple roles in the team that produces an ad) and allow us to calculate that, in this case,  15.1% of creators have played multiple roles.  The number of loops, edges that connect a node to itself, is 0. This is an artifact of our data, a 2-mode (bipartite) network in which which a node of one type can only be connected to a second node of the other type (Ads to Creators or vice-versa).

  The next set of numbers provide us with two common measures of network structure: Density and Degree. Density is the percentage of edges found in the data when compared to the total number of possible edges (n(n-1)/2) in a network with n nodes). As a result, Density tends to decline as network size increases. Degree, the number of neighboring nodes directly connected to the the node in question, is a more informative measure. The 4.1937158 reported here could, however, be misleading.  It tells us that, on average, every node in this 2-mode network is connected on average to between 4 and 5 nodes of the opposite type. It does not tell us either the average number of creators involved in producing an ad or the average number of ads produced by a single creator. To discover these facts requires further analysis.

  The last set of numbers tells us that, if this 2-mode network were represented by a matrix, the matrix would have 808 rows and 2661 columns. For mathematical purposes, networks are often represented as matrixes and methods from matrix algebra are used to analyze them. Here the rows are Ads, the columns Creators, and a non-zero number in a cell where a row and a column intersect indicates that there is a relationship between the Ad and Creator in question. In this case, the commands Net>Partition>2-mode and Info>Partition produce the following table.

Dimension: 3469
The lowest value:  1
The highest value: 2

Frequency distribution of cluster values:

Cluster     Freq   Freq%  CumFreq CumFreq% Representative
———————————————————————————————-
    1     808   23.2920     808   23.2920 AD1_06
    2     2661   76.7080     3469 100.0000 Yam342
———————————————————————————————-
  Sum     3469 100.0000

Here we find little that we do not already know. The table tells us that we are looking at partition that divides the total network into two clusters labeled 1 and 2 of which the first contains 808 members, the second 2661. Since these are the Ads and Creators, there is nothing new for us here.  Digging a bit deeper, we find that yellow and green are the default colors that Pajek assigns to clusters numbered 1 and 2. We also note the column headers:  Freq=frequency; Freq%=percentage of total; CumFreq=Cumulative Frequency; CumFreq%=cumulative percentage of total; and Representative is simply an identifier for a typical example of the cluster. With only two clusters to worry about, this may seem a lot of bother for no great reward. Suppose, however, that we change the partition by using Net>Partition>Degree>All and Info>Partition. Now the table that appears is the following

Dimension: 3469
The lowest value:  1
The highest value: 60

Frequency distribution of cluster values:

Cluster     Freq   Freq%  CumFreq CumFreq% Representative
———————————————————————————————-
    1     1261   36.3505     1261   36.3505 Tan1892
    2     623   17.9591     1884   54.3096 AD760_06
    3     255   7.3508     2139   61.6604 AD83_06
    4     277   7.9850     2416   69.6454 AD103_06
    5     229   6.6013     2645   76.2468 AD5_06
    6     183   5.2753     2828   81.5221 AD67_06
    7     99   2.8538     2927   84.3759 AD38_06
    8     104   2.9980     3031   87.3739 AD76_06
    9     76   2.1908     3107   89.5647 AD60_06
    10     54   1.5566     3161   91.1214 AD43_06
    11     39   1.1242     3200   92.2456 AD104_06
    12     50   1.4413     3250   93.6869 AD1_06
    13     32   0.9225     3282   94.6094 AD22_06
    14     21   0.6054     3303   95.2148 AD65_06
    15     27   0.7783     3330   95.9931 AD15_06
    16     16   0.4612     3346   96.4543 AD13_06
    17     25   0.7207     3371   97.1750 AD80_06
    18     16   0.4612     3387   97.6362 AD19_06
    19     12   0.3459     3399   97.9821 AD37_06
    20     17   0.4901     3416   98.4722 AD73_06
    21       7   0.2018     3423   98.6740 AD75_06
    22       3   0.0865     3426   98.7604 AD450_06
    23       5   0.1441     3431   98.9046 AD366_06
    24       6   0.1730     3437   99.0775 AD195_06
    25       6   0.1730     3443   99.2505 AD210_06
    26       3   0.0865     3446   99.3370 Ich674
    27       1   0.0288     3447   99.3658 Sas1520
    28       4   0.1153     3451   99.4811 AD448_06
    29       1   0.0288     3452   99.5099 Ter536
    31       6   0.1730     3458   99.6829 AD315_06
    32       1   0.0288     3459   99.7117 Yam342
    34       5   0.1441     3464   99.8559 AD311_06
    39       1   0.0288     3465   99.8847 Soe903
    42       1   0.0288     3466   99.9135 Oka258
    49       2   0.0577     3468   99.9712 Sas3
    60       1   0.0288     3469 100.0000 Saw8
———————————————————————————————-
  Sum     3469 100.0000

Here we learn that Degree, the number of immediate neighbors of nodes, ranges from 1 to 60. We may also note that the Representative nodes for the top four numbers are all Creators instead of Ads. We may also note the highly skewed distribution of degrees, from 1,261 nodes with only one immediate neighbor to only 1 node with 60 immediate neighbors. What could these observations means?

  The commands that generated the Degree partition also created a Degree vector. We have already noted that the cluster numbers in partitions are only labels for discrete categories. These can be handy for extracting subnetworks of nodes belonging to those categories but cannot, being only labels, be used in calculations. The numbers in vectors are real numbers (pun intended). They can be used in calculations. If we have already used Net>Partition>Degree>All as indicated above, Info>Vector produces the following table

Dimension: 3469
The lowest value:          0.0003
The highest value:          0.0173
——————————————————————-
Sum (all values):          4.1949

Arithmetic mean:            0.0012
Median:                  0.0006
Standard deviation:          0.0014
2.5% Quantile:            0.0003
5.0% Quantile:            0.0003
95.0% Quantile:            0.0040
97.5% Quantile:            0.0052

    Vector Values               Frequency     Freq%  CumFreq CumFreq%
——————————————————————————————————————————
(            ...        0.000]    1261   36.3505   1261   36.3505
(        0.000 ...        0.006]    2155   62.1216   3416   98.4722
(        0.006 ...        0.012]      49     1.4125   3465   99.8847
(        0.012 ...        0.017]      4     0.1153   3469 100.0000
——————————————————————————————————————————
  Total                           3469   100.0000

  Here we find the kinds of numbers used in statistical analysis: the range, mean, median, standard deviation, and the points one, two, three and four standard deviations from the mean. In the table the label of on the label on the leftmost column has changed from “Cluster” to “Vector Values,” and the numbers in the leftmost column have changed from integers used as labels to ranges defined in terms of real numbers.

  In this section, we have learned that Pajek can be used to dissect as well as analyze whole networks and begun to get a sense of how to think about the numbers that Pajek generates the diagrams with which we begin our explorations of our primary data: the 2-mode networks in which the members of winning teams and the ads they created are nodes and the roles that linked creators to ads the ties that connect them.

Analysis generates numbers, measures of either structural properties of whole networks or sub-networks considered as wholes or of the nodes and edges for which these networks are composed. Visualization draws pictures of networks, creating the network diagrams that are the first thing likely to come to mind when networks are mentioned. We must not forget, however, that these diagrams are only suggestive; they may stimulate thinking and suggest insights, but for reasons that will soon be apparent prove nothing about the data they display. Images can be misleading and are even easier to manipulate that statistics.

Consider, for example, Figure 1, the complete Ad-Creator network for 1981.
 
In this diagram, the network is displayed with the nodes (Ads or Creators) arranged in an oval and connected by a dense web of relationships, each one of which corresponds to a role connecting one Ad to one Creator.Yes, there are a lot of connections here, and the image suggests that all of the nodes are connected in one way or the other. This conclusion, however, is erroneous.

Figure 2 shows the same network; but now the network has been energized using the Fruchterman-Reingold algorithm that rearranges nodes in a way that separates unconnected parts of the network.

Now most of the nodes are spread around the oval on the outside, and only a relative handful are in the center where relationships connecting different segments of the oval intersect. Since, however, all of the nodes are of the same size and color, we see only the broadest outlines of the structure.

Figure 3 is the result of the Net>Partition> 2-mode command that creates a partition separating Ads and Creators and assigns different colors to them.

But this is only a bit more informative than Figure 2. Different partitions will tell us more. The Net>Components>Weak command creates a partition that divides the network into components, discrete subnetworks, such that members of any one subnetwork are connected with each other but not to members of any other component.  Figure 4 shows the result of applying this command to the 1981 Ad-Creator network.

The colors correspond to a structural attribute, the node’s membership in one and only one component. In this diagram, the vast majority of nodes, the ones colored yellow, belong to a single giant component. This network contains a total of 73 components; but the rest are very small, a result consistent consistent with basic theorems in network analysis math (see “Components and Bicomponents”).

Suppose, however, that we take a look at other attributes, taken from our data instead of structural analysis. Figure 5 shows the network with the nodes colored to represent the agency that produced the Ad (Yellow=Dentsu, Green=Hakuhodo, Dark Blue=Other) and to differentiate the Creators (Light Blue=Creator) from the Ads.

We may note how Ads produced by Hakuhodo (Green) dominate the center of the diagram where relationships connecting the whole intersect, and Ads produced by Other agencies (Dark Blue) far outnumber those produced by both Hakuhodo and Dentsu. Analysis of other networks will show that only the 1981 network exhibits these features.

Figure 6 shows the same network. Once gain Light Blue distinguishes the Creators from the Ads. But now the other colors correspond to media (Yellow=TV, Green=Radio, Pink=Newspapers, Red=Magazines).

Here we observe that the number of TV commercials is relatively small and the number of magazine Ads large. And the cluster at the center of the diagram may be a print campaign, combining newspaper and magazine Ads but including no TV or radio. Again, analysis of other networks will show that only the 1981 network exhibits these features.

In later sections, we will employ other features of Pajek to enrich our visualizations of the networks we are studying. But when I say now that Pajek functions as a sociological macroscope (the functional equivalent of the biologist’s microscope), my meaning should be clear. Already we can see structures emerging, and like the biologist who stains his slides, we, too, can use color to highlight questions of interest. But we, too, must be careful when we draw conclusions from what we see. Visualization must be followed by analysis based on the numbers that Pajek generates for us as it creates network diagrams, and these in turn must be considered in light of the larger context from which those numbers are extracted.

Pragmatically, SCA aimed to identify the levers required by social engineering. Conceptually, it saw society as the outcome of forces operating in a way fundamentally similar to those envisioned by physicists in classical mechanics. These forces impinged on human objects (individuals or groups), and compelled them to move in particular directions. It was imagined, for example, that a particular combination of forces labeled “race,” “gender,” and “income” would necessarily determine the direction of a woman’s life. Those who objected to that direction could, then, alter it by changing one or more of the forces in question.

The problem was how to identify the forces and measure the weight with which they impinged on the human object. The favored method was a large-N statistical analysis of randomly sampled data whose output is epitomized by regression lines, i.e. vectors,a.k.a., directions, straight lines from point A to point B, constructed by optimizing the weightings of the forces (other vectors) supposed to impinge upon them. Since the subjects who answered the survey questions were usually individuals, this method was also highly compatible with envisioning an ideal outcome in which democracy (votes by individuals) and market forces (consumer choices by individuals) would aggregate to form the forces that would, via negative feedback, keep that society in stable equilibrium.

As critics never tire of pointing out, however, this approach not only ignores the agency, the human freedom to choose, of those whose behavior is shaped by the forces in question. Reality is messy and the results of research conducted along these lines tend to be neither clean nor pretty. The heart of the critique is captured in the quote with which Lin Freeman begins his history of social network analysis. Attributed to Columbia sociologist Alan Barton and written in 1968, it reads as follows.

For the last thirty years, empirical social research has been dominated by the sample survey. But as usually practiced, using random sampling of individuals, the survey is a sociological meat grinder, tearing the individual from his social context and guaranteeing that nobody in the study interacts with anyone else in it. It is a little like a biologist putting his experimental animals through a hamburger machine and looking at every hundredth cell through a microscope; anatomy and physiology get lost, structure and function disappear, and one is left with cell biology…If our aim is to understand people’s behavior rather than simply to record it, we want to know about primary groups, neighborhoods, organizations, social circles, and communities; about interaction, communication, role expectations, and social control. (Freeman, 2004:1)

Without the sociological turn to which Barton points in his conclusion, we are left with trying to understand what goes on inside the human cells of the social organism. When that is conceived in terms of subjective feeling, meaning and choice, we are left with either the speculations of consumer research or, in academia, the interpretive/postmodern alternatives previously mentioned above.  With that sociological turn, however, the way remains open for scientific research that is not necessarily constrained by SCA assumptions. Social network analysis is one prime example; agent-based modeling may be another. Classical forms of ethnography and history, which balance interpretation from the subject’s point of view with material conditions, also have much to contribute.

What is it exactly, however, that social network analysis brings to the table? According to Freeman, social network analysis is a combination of

1. Structural intuition based on ties linking social actors
2. Systematic empirical data
3. Graphic imagery, and
4. Use of mathematical and computational models. (Freeman, 2004:3)

In this study, the structural intuitions that motivate the research are derived from a common sense representation of the Japanese advertising industry as a range of mountains, three very large and others smaller, fading into foothills. That the mountains form a range is important. The feet of the mountains are joined. How separate the peaks actually are or, alternatively, represent the joint outcome of inferred tectonic processes become interesting questions. It is ties formed by social actors, advertising creatives who work together in teams, that suggest an approach to finding some answers. The systematic empirical data are provided by the credits in the TCC annual; heavy use is made of graphic imagery, especially network diagrams; and the mathematical and computational models incorporated in Pajek not only generate the diagrams, they also suggest a variety of ways to parse and explore the data.

A primary focus on relationships, ties between social actors, instead of actor attributes, is a feature cited by every introduction to social network analysis. Wasserman and Faust note four additional features frequently found in social network analyses.

•   Actors and their actions are viewed as interdependent rather than independent, autonomous units.
•   Relational ties (linkages) between actors are channels for transfer or ‘flow’ of resources (either material or nonmaterial).
•   Network models focusing on individuals view the network structural environment as providing opportunities for or constraints on individual action.
•   Network models conceptualize structure (social, economic, political, and so forth) as lasting patterns of relations among actors.(Wasserman and Faust, 2005 [1994]:4).

Thus, for example, in this study the actors, the advertising creatives, work together in teams. Team members bring different skill sets to the table, all of which are required to produce award-winning advertising. They are certainly interdependent. Working together in teams involves flows, minimally the exchange of ideas required to reach consensus on a proposal and produce the ad in question. Working in a team with other talented people presents opportunities; working with winning teams is a way to advance a career. Working in teams also involves constraints. Advertising creatives are notoriously independent thinkers, often prima donnas. But to work together successfully requires give and take and knowing how far to push without pushing too hard.

It is, however, an interesting question whether the actors who make up the teams form lasting relationships. Teams are formed for specific projects. Team members may go their separate ways once a project is over. Our data show us only that the members of a team worked together on one or more winning ads, and the networks constructed by contest judges decisions are only subsets of larger professional networks.

Mark Newman reminds us that, mathematically speaking, a network is simply a collection of points joined together in pairs by lines.  (2010:1) In the case of social networks, the points represent actors, the lines the social ties between them. But precisely the same sort of representation can be used for all sorts of networks: the Internet and Worldwide Web, transport and power grids, and protein cascades in cell biology are frequently cited examples. Why, then, a reader might ask, turn to network analysis for help in understanding systems in which so much else are involved? Why, in particular, use network analysis to study people, whose feelings, thoughts and meanings are not represented in network diagrams? One answer is that all these can added to the framework that network analysis provides, as the overall plan for this research suggests. Newman provides another, and to me compelling, answer.

Scientists in a wide variety of fields have, over the years, developed an extensive set of tools?\mathematical, computational, and statistical?\for analyzing, modeling and understanding networks?c. Thus if there is a system you are interested in, and it can usefully be represented as a network, then there are hundreds of different tools out there, already developed and well understood, that you can immediately apply to the analysis of your system. Certainly not all of them will give useful results?\which measurements or calculations are useful for a particular system depends on what the system is and does and on what specific questions you are trying to answer about it. Still, if you have a well-posed question about a networked system there will, in many cases, already be a tool available that will help you address it. (2010: 2-3)

For this particular project, Pajek provides a handy toolkit in which many useful tools can be found. There is, however, one additional set of issues to consider about the way in which those tools are used.

Wasserman and Faust note that, “the methods of social network analysis can be used in two ways, as formal descriptions or in model and theory evaluation and testing.” (2005[1994]:5). Here I would like to suggest a third. The formal descriptions and network diagrams generated by Pajek and similar software can function, in effect, as the social scientist’s microscope. Like the biologist’s microscope, they reveal fine details of structure that are normally invisible or only dimly perceived. This simile gains added force from network visualization software’s use of color, shape and size to code network node attributes (as, for example, assigning a color that indicates the lead agency in charge of producing an ad or making actors with more direct connections to others visibly larger than others). These capabilities allow this network analyst to follow the biologist’s example and stain his slides to highlight structures of particular interest. Other tools make it possible to dissect complex structures and extract portions of them that are simpler to understand. The formal descriptions provide a richly detailed structural framework on which to hang other considerations, from personal perspectives to material conditions.

This point, however, marks the boundary of this study. Model and theory evaluation and testing require mathematical and computational expertise that others will have to supply.

Step one was to read some popular books on the subject: Duncan Watts (2003) Six Degrees: The Science of a Connected Age and Albert-Lazlo Barabasi (2003) Linked: How Everything is Connected to Everything Else, for example. These books taught me that social network analysis is a subset of a larger field called network analysis, whose mathematical principles apply to transportation and power grids, the Internet and protein cascades as well as social networks. That was exciting. But while books in this genre were good for pumping up my interest, they didn’t teach me how to do network analysis.

An old friend, sociologist James Ennis put me on to UCINET, academia’s most widely used social network analysis software, and searching the Web to learn more about that program brought me to Robert Hanneman and Mark Riddle (2005) Introduction to Social Network Methods (available online at http://www.faculty.ucr.edu/~hanneman/nettext/). This text whetted my appetite but was plainly designed for students with professors handy to answer their questions. As an independent scholar, I was, or so it felt to me, left working out too much for myself by trial-and-error.

Pursuing other leads brought me to Stanley Wasserman and Katherine Faust (2005 [1994]), Social Network Analysis: Methods and Applications, a frequently cited work that provides a very thorough grounding in the mathematics of network analysis and its application to social networks?\and is tough going for a newcomer who is still trying to resurrect what he learned in undergraduate courses in logic, calculus and probability. I ran into a similar problem when I purchased a copy of Mark Newman, Albert-Lazlo Barabasi, and Duncan J. Watts (2006) The Structure and Dynamics of Networks.This collection of classic papers is a deep introduction to the history of network analysis and nicely complements Lin Freeman (2004) The Development of Social Network Analysis: A Study in the Sociology of Science, which provides a wealth of the human detail concealed in more abstract introductions to the field. That said, it, too, is tough sledding for a beginner unfamiliar with the relevant mathematics.

Today getting started is easier than it was even just three years ago. Newcomers to the field can start with Alexandra Marin and Barry Wellman (2009) “Social Network Analysis: An Introduction” forthcoming in the SAGE Handbook of Social Network Analysis. The year 2010 also saw the publication of three new textbooks that target advanced undergraduates and graduate students taking a first or second course in network analysis: David Easley and Jon Kleinberg (2010) Networks, Crowds, and Markets: Reasoning About a Highly Connected World, Matthew O. Jackson (2010) Social and Economic Networks, and M. E. J (Mark) Newman (2010) Networks: An Introduction. All are written by masters in the field and are relatively accessible to readers whose math is at my recovering undergraduate level. I would still, however, need to learn not only basic network concepts but also how to use a software package to analyze my data.

The book that turned my life around and made this research possible is W. de Nooy, A. Mrvar, and V. Batagelj (2005) Exploratory Social Network Analysis with Pajek,  a book that lives up to the promise of its blurb and is, in fact, ” the first textbook on social network analysis integrating theory, applications, and professional software.” The concepts used in my research were, if not first encountered, first understood, at least in part, by working through this book. For an independent scholar there was also the added advantage that Pajek, written and updated by Andrew Mrvar and Vladimir Batagelj and their colleagues at the University of Ljubljana, is freeware and easily handles networks with thousands of nodes.

But why these three tables, and how do they function? The following discussion is a bit technical but deserves close attention from readers who wish to know how the analysis reported here was done as well as the conclusions drawn from it.  It may also be useful for those contemplating similar research to know that working out these details took me several months of intermittent hacking before I got them clear.

In this type of database, the data are stored in fields, with the fields referring to a single case combined in a single record. To readers familiar with spreadsheets, it suffices to say that if a table is represented in spreadsheet format the fields are the the columns; the cases are the rows. But why, in this database are three separate tables required?

It is a basic principle of database design that there are three and only three fundamental relationships between the data contained in separate fields: one-to-one, one-to-many, or many-to-many. If the relationship is one-to-one, the two fields can be combined in a single table. If the relationship is one-to-many, two tables are required, with a unique identifier that appears in both tables used to link one to the other. If the relationship is many-to-many, three tables are required. Two of the tables have a one-to-many relationship to the third, called the join table, that provides a bridge between them. To see how this works in practice, we now take a closer look at the data contained in the Ad, Creator, and Record tables.

Each ad that appears in the TCC annual is a unique case, with several attributes, each of which stands in a one-to-one relationship with the ad in question. Each ad has only one sponsor, one lead agency, one industry category, one year and one medium. These basic items all appear with the ad in its record in the Ad table. Calculation fields (a special kind of field that displays the result of a calculation) are used to code the raw data in the basic fields; so that, for example, ad agencies can be coded as Dentsu=1, Hakuhodo=2, ADK=3, and Other=4.

Each creator has a personal name and surname. But here we have a problem. Different individuals may have the same names. The solution is to assign each creator a unique serial number. A calculation field is used to combine the first three letters of the surname with the serial number to create the name used in the social network analysis. Thus, for example, a copywriter named Maki Jun (where “Maki” is the surname and “Jun” the personal name), whose unique serial number is 65, becomes Mak65.

While calculation fields are used for coding information in both the Ad and Creator tables, the basic one-to-one relationship between the fields in each table remains undisturbed. In the Role table, which functions as a join table and provides links between ads and creators, things are a bit more complicated.

The problem here is that creators and ads have many-to-many relationships. The same creator may be credited with several different roles in the production of an ad: the same individual may, for example, have played three roles, creative director, copywriter, and planner. In addition, two or more individuals may be credited with the same role: thus, for example, multiple copywriters may have worked on the same ad. Records in the Role table each contain three vital pieces of information: a unique identifier for an ad; a unique identifier for a creator; and a category label for the type of role in question: copywriter, art director or film director, for example.The unique identifiers are automatically generated serial numbers assigned when new ads or creators are added to the database, and the pair of unique identifiers specify precisely the particular role in question. It is the roles specified in this way, as unique combinations of ad and creator that become the links analyzed by the social network analysis software.

In sum, the ads and creators are nodes in the networks analyzed. The roles are the links which connect them.