Digital media influences culture — and it’s influenced by culture in turn. Culture matters in business: Facebook just spent an astonishing $19 billion to acquire WhatsApp because of WhatsApp’s international presence. Culture also matters politically: Turkey’s leader recently made it the latest country to try blocking Twitter. How can we use data to understand culture’s impact on digital media adoption and usage? How can we even measure culture in the first place?
Dimensions of Culture, And How They’re Being Measured
Some cross-cultural elements are easy to see. For example, when a brand or platform expands into a new country, their stuff should obviously be translated. Sometimes marketers use basic analytical tools to cut words that are controversial in a given culture, like the Chinese localization service Kawo, which screens English tweets for words that would be sensitive in China before any actual translation is done.
Are there more complex tools that use language to measure or compare cultures? I figured that sentiment analysts might have some ideas, so I spoke to Seth Grimes, an industry analyst and the organizer of Sentiment Analysis Symposium. “In general social media analytics,” Grimes said, “the question comes up: How do you handle people talking in dozens of languages? The first answer that people come up with is machine translation: Translate everything to a single language and do the analysis there. But there’s a lot of idiom that just doesn’t translate well. For instance, there are sayings like ‘Rome wasn’t built in a day.’ So the prevailing case is that people are building sentiment analysis language-by-language.”
This separation and other factors make it hard to compare sentiment analyses across languages. For further technical reading, I found a 2013 research paper that reviews previous cross-language sentiment analysis strategies, discusses their drawbacks, and describes testing one potential approach (full PDF; Hogenboom et al). The researchers’ main discovery seems to be that cross-language sentiment analysis is neither obvious nor easy.
Individualism and Collectivism
There are plenty of cultural attitudes that are harder to spot than a change in language, but are important nonetheless. One much-discussed pair of cultural factors is “individualism” vs. “collectivism,” which are often presented on a spectrum. Collectivist cultures emphasize harmony, interdependence, and group identity. Individualist cultures enjoy talking about heroes, personal achievement, and “being yourself.” The USA is generally considered an individualist culture.
Perhaps predictably, collectivist cultures generally use social media more, according to an upcoming study by Adam Acar (who was kind enough to send me an advance copy). Acar also noted larger usage patterns, suggesting that it might be true that “rich countries tend to be more individualistic and autonomous, thus use the internet for individual purposes such as entertainment or information browsing. People from developing countries, however, tend to be more collectivistic and use social media to get emotional and social support.”
There are certainly differences between the types of messages that individualist marketers put on social media, as opposed to collectivist marketers. For instance, a detailed 2012 study found that collectivist marketers on the Chinese social platform Renren are more likely to say that their product is very popular (abstract; Tsai and Men). They’re also much more likely to appeal to group identities, like by addressing messages to “business professionals” or “college freshmen.” In contrast, individualist American marketers on Facebook talk about uniqueness and independence, like a sportswear company asking “What is your signature style?”
Studies that measure cultural attitudes usually rely on people to take specific content units (such as tweets or Facebook posts) and then give them a numerical rating for a specific cultural dimension. So, for instance, a person working on a study might take 100 tweets and manually rate each tweet as individualist vs. collectivist on a 1-5 scale. Then those numbers are batched and crunched by the researchers. It might be possible to use machine learning to develop a sentiment analysis tool that detects cultural attitudes, but no one appears to have done that yet.
Another useful spectrum for looking at cultural attitudes is “high context” vs. “low context.” (High vs. low context are also sometimes called “indirect communication” vs. “direct communication.”) A high-context culture is one where more things are left unsaid; people make more assumptions about each other based on clothes, postures, or even long silences. In low-context cultures, people make fewer assumptions and use more precise language. They ask a lot of questions and don’t mind being explicit. The USA is generally considered a low-context culture.
One 2010 study found that business websites from low-context cultures are more accessible through search engines (abstract; Usunier and Roulin). In other words, people from low-context cultural backgrounds are better at optimizing their sites for search than people from high-context cultures. This makes sense, since search engines originally emerged from low-context cultures; the researchers noted that countries such as the USA, Australia, Israel, and Switzerland are lower-context. (Higher-context countries include Egypt, Russia, China, Singapore, and Kuwait.)
The 2012 study that I mentioned earlier, which looked at individualist vs. collectivist marketers, also examined how other cultural factors influence both features and content on Facebook as opposed to the Chinese social platform Renren (abstract; Tsai and Men). It found that Renren hosted more high-context marketing appeals than Facebook. These high-context appeals include metaphors, as well as “references that were not relevant to the brand, company, or product category.” For instance, the researchers give the example of a Chinese eyewear brand that encouraged its followers to drive home safely after a long day at work — which is a nice thing to say, but is not relevant to eyewear.
Visual elements like the style and colors of pictures on social media have been linked to culture. For instance, a 2012 study analyzed Instagram pictures from Tokyo and New York using digital image processing software (full PDF; Hochman and Schwartz). One of their findings was that Tokyo images had more red-yellow hues while New York photos were bluer.
Of course, communication styles matter for design, too. Studies from over ten years ago were already demonstrating that personal websites from high-context cultures such as Korea are more likely to use illustrations to represent authors, rather than the photographs used by authors of US sites (full PDF; Kim and Papacharissi). This pattern applies on social media platforms, too, according to the 2012 Tsai and Men study. They found that high-context Chinese marketers used more “ambiguous” visuals such as illustrations, while low-context USA marketers used direct visuals like brand logos.
It’s important to note that there are a lot of dimensions for culture, and scholars don’t agree about all of them. I’m just scratching the surface by mentioning language, design, individualism, collectivism, and high vs. low context.
Social Media Metrics
As I mentioned, much of the available research about cultural differences in social media has used people — rather than tools — to rate how cultural values are expressed. Researchers use software when they can, but that’s a lot more feasible when you’re seeking colors or specific words than it is when you look for collectivism.
In general social media marketing, people often use the inherent metrics of social media to measure a campaign’s impact: Likes, shares, retweets, hashtag usage, and others. Could those metrics be used cross-culturally? I found a 2011 study that measured the efficiency of international brands’ Twitter strategies by looking at the number of followers per tweet (abstract; Burton and Soboleva). It concluded that Microsoft’s USA Twitter account was more efficient than Microsoft’s Australia Twitter account because the USA account had more followers per tweet.
This approach can be useful, but one problem is that the meaning and usage of a metric may change from culture to culture. For instance, I already mentioned a 2013 paper about cross-language sentiment analysis; those researchers tried to develop a standard cross-language measurement by using numerical star ratings in movie reviews, but they had trouble with consistency (full PDF; Hogenboom et al). We can’t assume that follows, shares, and so on mean the same thing to people in different cultural environments.
Usage of digital media can change a lot across cultures, and we’re just starting to pin down some of those patterns. This post focused on measuring language, design, and cultural attitudes; my next post will give some cross-cultural case studies, with an emphasis on commerce and censorship.