Small and Large Hashtags: Parallels to Consumer Behaviour and Differing Double Jeopardy Effects in the Twittersphere.
*Cullen Habel. The University of Adelaide. firstname.lastname@example.org
Jean-Eric Pelet. Université de Nantes. email@example.com
Keywords: Twitter, Metrics, Double Jeopardy, Hashtag, Social Media
* Presenting author
Affiliate groups in Twitter are represented by hashtags, with each tweeter typically only posting one or two tweets to the hashtag in a given period. This paper reviews an archive of 100,000 tweets over 30 hashtags to find support for the propositions of large differences in market share and number of tweeters per hashtag, but small differences in the average number of tweets per tweeter. The low end shows an inverse double jeopardy line, where the smallest hashtags tend to be dominated by a heavy tweeter trying to drive discussion. Mature affiliate groups such as #qanda sit on the traditional DJ line. We conclude that reach is important – broad contact is preferable to dozens of tweets sent to a small follower base.
This paper addresses a question that is familiar to practitioners – should they pursue reach or frequency? Will a better “buzz” be created by engaging very broadly, or by building a close community of strong advocates? It does this in the context of the Twitter platform and demonstrates that while the medium may look different, the pursuit of reach is still important.
Twitter enjoys a global presence of 200 million users (Shiels, 2011) and is considered to have had a major influence on both the US elections and the Iranian elections of 2008 and 2009 (Shiels, 2011), whilst a “Mashable” poll indicated that Twitter broke news of Osama Bin Laden’s death to over 25% of their respondents with television next at 18% (Parr, 2011).
As a communications platform, Twitter is only slightly younger than Facebook, starting in 2007 compared to Facebook’s 2004 (Carlson, 2010). Twitter was conceived as a network of SMS messages – similar in some ways to the “news feed” function of Facebook. Twitter represents an alternative marketing communication and networking channel with a range of practical applications, including:
- Extending reach of existing blog or other communications as with mashable.com, an online communications blog.
- Announcing events and deals – Dell and Amazon, as well as group wine buying sit @vinomofo
- Improve frequency of updating websites and blogs – @theaustralian (newspaper) has a strong Twitter feed for this
- Building consensus and communities of supporters – @theqwoffboys have included twitter as a medium to build their strong networks
- Building word of mouth “buzz” around ones product or brand – the city of Adelaide benefits from the services of @adelaidetweet
- Updating breaking news at conferences or events - @GreatWineAdv is the “Great Australian Wine Adventure”
- Updating networks for personal branding – Stephen Fry is one of the strongest on Twitter
(Source: Handley, 2007)
Given the maturity and the continued growth of the platform there has been – somewhat surprisingly - little scholarly work on the patterns of communication within this medium from a marketing perspective.
With a key marketing application of Twitter as an outreach medium, as indicated in points 1, 4 and 5 above, this paper reports upon an exploratory quantitative investigation into the nature of “buzz” creation and relates it to the well known “double jeopardy” pattern of consumer brand selection.
Hashtags indicate communities of interest
Whist the “follow” and “news feed” functions of Twitter tend to mimic those of Facebook; it is the use of hashtags that indicate self organising affiliate groups, and allow users to “buy into” a given area. A user may include the term #justinbieber in their message to signify their engagement with the “Justin Bieber” meme, and other users may choose to include the same term if they – too – are speaking to that meme. Hashtags may be frivolous, or more serious as shown by list given in table 1.
Rather than a personal stream, users may then choose to search on the hashtag term and view all messages (from followers/followeds or not) regarding that term. A user watching current affair program “The Insiders” is able to follow an underground, global, real time conversation simply by watching the #insiders stream on their PC, iPad or smartphone.
How strong is an affiliate group?
In the pursuit of a large, active group of tweeters affiliating with their hashtag, a marketer should be concerned with two measures – how many tweeters have engaged and the average number of tweets per tweeter. This has a distinct parallel with previously identified consumer goods patterns (Uncles et al., 1995) and allows us to present a set of simple propositions for our twitter study:
P1: Large differences occur in market shares (hashtag daily tweets)
P2: Small differences in average purchases per buyer between brands (tweet rates)
P3: Large differences in the number of buyers (tweeters)
P4: Double Jeopardy Effects – tweeters and tweet rates show a positive correlation across the hashtags.
In much the same way as consumers have a choice to buy multiple brands over a period of consumption, tweeters have the choice to engage with a broad range of topics on twitter – as represented by various hashtags. This exploratory paper has taken a very broad range of these hashtags and they are not meant to represent brands or product classes, simply to gain an insight to overall tweeter behaviour.
During the month of June, 2011, we observed the tweeting behaviour around some thirty hashtags. This involved archiving tweets for a given period using a freeware product (The Archivist) provided by a division of Microsoft. This data collection software makes use of the twitter search function, saving the username, plus the time and content of each tweet that used a specified hashtag. This allowed us to tabulate the average tweets per day made using a given hashtag, as well as the number of engaged tweeters and average number of tweets per engaged tweeter. There appeared to be a natural break between hashtags – they either enjoyed over 1,000 tweets per day (on average) or 500 or less. For analysis later in this paper this natural break was used as a way of defining small and large hashtags.
Results and Discussion
Table 1 below gives a sample of the data, from the extremely active (if frivolous) meme where people might say “#dontlookatmeif I haven’t had my first coffee for the day” to tweeters watching a television show with “Politest #qanda panel ever.”, and “go to #wineFORUM for great wine news”.
Only five small and five large hashtags are shown below, for reasons of brevity. Readers are welcome to contact the author for the exhaustive list of the 30 or so.
It must be noted that these observations are not for the same length of time, or for the same dates, or even the same pool of tweeters. It is therefore not strictly correct to refer to “market share” when comparing hashtags to each other. Nevertheless, the tweets per day represent relative activity within the particular hashtag, some “share” of the twitter attention being paid to a given meme.
Even leaving aside the extreme first value, it can be seen that the rate of tweets per day vary between around 10,000 and perhaps around a hundred (sometimes down to 50 or 30). Thus while accepting the fact that measures are not directly comparable, proposition 1 is supported. This can also be seen in the x-axes of figure 1.
When we consider the average number of tweets per tweeter see that even within the restricted sample of table 1 the average number of tweets per tweeter varies between 1.2 and 3.9, with an outlier of 5.6. That is essentially a 1:4 variation from largest to smallest, compared to very large variation that occurs in the tweets per day column, so proposition 2 is supported. Graphically this appears in the left hand y-axis of figure 1.
When we consider the number of tweeters in each twitstream it can be seen that they vary from a hundred or less, into the thousands, as shown in the y-axis of the right hand graph of figure 1. Thus P3 is supported.
Double Jeopardy in the Twitterverse
Proposition 4 deals with the phenomenon of double jeopardy (DJ). In a marketing sense, DJ means that lower market share brands are penalised twice – they have far fewer buyers and the buyers they have tend to buy them less often (Ehrenberg et al., 1990). The Twitter context tends to parallel the original observations of McPhee (1963) with Hollywood actors, comic strips and radio presenters. A DJ line is an x-y plot of the number of buyers of a brand against the number of times the brand is bought (Allsopp and Jarvis, 2003) and tends to be an upward sloping line with some curvature (Habel and Rungie, 2005).
A DJ line in the Twitter world is a plot of tweeters vs tweets for the various hashtags. Such a plot for the entire set of this data showed almost no relationship (R2=0.04) with a slight downward slope. This amounts to a rejection of proposition 4. Double Jeopardy does not appear to apply when the entire sample of twitstreams are considered. This is worthy of further consideration.
Different DJ lines for the small and the large hashtags
This contrary picture to standard DJ can be explained by the way in which small hashtags (<1000 tweets per day) behave differently to the larger ones (>1000) as shown in figure 2.
For the large hashtags the standard DJ pattern applies – upward sloping line with an R2 around 0.3. It is with the small hashtags that the opposite occurs – the fewer tweeters there are the more tweets – on average – they tend to make.
Driving communities vs driven communities.Closer investigation of the tweeter behaviour shows a great many of the smaller hashtags being dominated by a small number of highly active members. In one case (a wine forum) four tweeters accounted for 65% of all tweets - actively working to generate interest. This means that the chart on the LHS of figure 2 has a great many of those small brands showing high average tweet rates per tweeter.
By contrast with a large, a mature hashtag such as for a television show (in this case #qanda) there is no dominant tweeter. Some are heavier, but there is no one tweeter dominating the discussion. Past some critical point the community gains a life of its own and does not need to be driven by PR consultants and forum owners. This is also an indication that external drivers – such as being a television show of its own (or some global meme) is a great assistance to an affiliate group.
We argue that this is the cause for the rejection of proposition 4, and the reason for the differing double jeopardy effects when we consider the small and the large brands – it is an artefact of the small communities being driven by a number of highly active members.
Conclusion and Marketing Implications
Whilst the propositions of this paper are not – in themselves – controversial they have not been examined in the context of this new medium of twitter. Propositions 1, 2 and 3 draw a clean parallel to brand choice in consumer goods marketing – hashtags vary far more in the number of people engaging with them than they do in the level of engagement.
For a marketer or communications professional attempting to generate activity around their topic of interest (their own hashtag) this means that reach is important. The overwhelming nature of Twitter is one of tweeters engaging briefly with a range of ideas as they go use the medium. The strong hashtags have thousands of tweeters making around three tweets.
Marketers and communicators should therefore work with the ephemeral nature of the medium. The gaining of a single retweet from a tweeter with thousands of followers is likely to promote your hashtag far more effectively than spamming your own five hundred with dozens of messages.
Limitations and Further Research
This has been in its very nature a coarse analysis. We have deliberately looked past the nature of the tweet content and across a range of hashtags in an effort to understand consumer behaviour patterns in this medium. Whilst the overarching patterns of double jeopardy have been established (and exceptions identified) there is ample scope to fine tune this analysis to consider what is being written in the tweet, or the more general categories of hashtag.
Beyond the coarse analysis of DJ lines and percentage of tweets, future research should include Pareto analyses – the percentage of tweets accounted for by the heaviest 20% of tweeters (e.g. Anschuetz, 1997). Furthermore, with such a skewed distribution of tweet rates (many low level tweeters and a few heavy tweeters) a range of consumer behaviour models such as the Negative Binomial Distribution (Ehrenberg, 1959) would take this to the next level.
Allsopp, J. and Jarvis, W. (2003). Contrasting the double jeopardy line with a multi-brand market. Australian & New Zealand Marketing Academy Conference. Adelaide.
Anschuetz, N., 1997. Profiting from the '80-20 rule of thumb'. Journal of Advertising Research 37 (6), 51-56.
Carlson, N. (2010). "At Last -- The Full Story Of How Facebook Was Founded." Business Insider Retrieved 22/6/2011, from http://www.businessinsider.com/how-facebook-was-founded-2010-3#we-can-talk-about-that-after-i-get-all-the-basic-functionality-up-tomorrow-night-1#ixzz1PyBBUZ4V.
Ehrenberg, A.S.C., 1959. The Pattern of Consumer Purchases. Applied Statistics 8 (1), 26-41.
Ehrenberg, A.S.C., Goodhardt, G. and Barwise, T.P., 1990. Double Jeopardy Revisited. Journal of Marketing 54 (July), 82-91.
Habel, C. and Rungie, C., 2005. Drawing a double jeopardy line. Marketing Bulletin 16 (1 - 10.
Handley, A. (2007). "7 Ways Marketers Can Use Twitter." from http://www.mpdailyfix.com/7-ways-marketers-can-use-twitter/.
McPhee, W.N. (1963). Formal Theories of Mass Behaviour. New York, The Free Press of Glencoe.
Parr, B. (2011). "How Did You Hear About Osama Bin Laden’s Death? [POLL]." Retrieved 4/10/2011, 2011, from http://mashable.com/2011/05/02/osama-bin-laden-death/.
Shiels, M. (2011). "Twitter celebrates its fifth birthday." BBC News Business Retrieved 22/6/2011, from http://www.bbc.co.uk/news/technology-12805216.
Shiels, M. (2011). "Twitter co-founder Jack Dorsey rejoins company." BBC News Business Retrieved 22/6/2011, from http://www.bbc.co.uk/news/business-12889048.
Uncles, M., Ehrenberg, A. and Hammond, K., 1995. Patterns of Buyer Behavior: Regularities, Models, and Extensions. Marketing Science 14 (3, Part 2 of 2), G61-G70.