For this research I was interested in taking a closer look at how South Africans navigate Twitter. I was particularly interested in whether there are any topics that South Africans don’t want to tweet about (topics which they self-censor) and who they imagine to be listening when they do tweet (the imagined audience). I also wanted to establish how users interact with follower metrics, as these have played such a significant role in the framing of Twitter as a social medium.
I employed a mixture of quantitative and qualitative research methods. According to Marres (2013) digital technology inaugurates an age of methodological innovation, as new technologies for data collection, analysis and visualization enable the further elaboration of existing methods and the development of new ones. Digital technology aided me greatly in the collection of data. I employed the following methods:
- Online surveys
- In-depth interviews
- Questions on Twitter
- An ethnography of Twitter
Research was conducted during a couple of weeks in July and August 2013.
1. Online survey
The Internet has made it much easier to distribute surveys and get more responses from a wider demographic. Internet surveys allow answers to be immediately stored in a computer database where they are ready for further processing.
I created a survey using Google Forms. This is a free tool that is very easy to use and customize. Responses are automatically stored in a Google Document and can be easily downloaded in various formats such as Excel or .CSV files. Other than with a tool such as SurveyMonkey, there is no limit to the number of questions one can ask or the number of responses that can be received. One can also embed Google Forms into a website using simple code.
A mixture of open-ended and closed questions was used. The survey was purposely kept short in order to increase the likelihood of responses. The survey initially included more open-ended questions, but after pretesting the questionnaire I decided to convert some of them to closed questions in order to make the completion process shorter.
The open-ended questions were used to give respondents enough space to express themselves and so that I could explore ideas that would otherwise not be aired. In order to keep the survey short, closed questions in the form of multiple choice questions were also included. The options for the multiple choice questions were chosen based on numerous discussions with Twitter users, online research and personal experience. Keeping in the style of Twitter, questions were kept short and to the point. Questions included:
- What won’t you tweet about? And Why?
- What are the main reasons why you unfollow people?
- Without looking at your follower list, who do you think follows you? Who reads your tweets?
- Go to www.twitter.com and scroll through your follower list. Does this correspond to what kinds of people you thought were following you?
- How often do you look at your follower list?
The questionnaire also contained some questions regarding demographics (gender, race, year of birth, occupation, language, Twitter handle). None of the questions were mandatory. This was done in order to increase the likelihood of responses and to encourage people to respond honestly and freely. The only question that was mandatory was the question concerning privacy at the end of the questionnaire. Users were asked whether they wanted to submit their answers anonymously or whether they consented to their Twitter handle being used in research results.
As I only wanted responses from South Africans on Twitter, I manually identified respondents and contacted them to complete the survey. I started with people in my follower network and other South Africans that I know use Twitter. I tweeted the link (from the @TweetMzansi account and my own account) and emailed it to a few people. It was important for me to do the bulk of the marketing on the Twitter platform, as these were the users that I was looking for. As soon as the @TweetMzansi account started gaining traction, I started sending Direct Messages to followers of the account to invite them to complete the survey. Followers were only contacted once and great care was taken not to spam people.
The survey was embedded into a responsive website so that it could adapt to any screen size. Respondents could thus complete the survey on any device at a convenient time.
The data was exported and manually coded and analyzed using Excel and Google Refine.
I was also interested in how people with bigger followings navigate Twitter in South Africa. These people are mostly well known people who are harder to contact on Twitter and who have limited time to respond to tweets and questionnaires. I decided to conduct some interviews, because these are generally easier and more convenient for the respondent.
Another reason why I chose to do conduct interviews was because I thought that they would complement my survey results and perhaps shed some more light or expand on some of my findings. Interviews provide more flexibility than a survey, one can skip irrelevant questions, ask unplanned questions, change the order of questions, and one can ask for clarification.
All interviews were conducted via Skype calls and email and were recorded on my iPhone and transcribed in full.
Respondents included well known politicians, journalists, a CEO of a bank, the former governor of the South African Reserve Bank, the editor of a local newspaper, and outspoken Twitter users with very large followings.
Questions asked included:
- What won’t you tweet about? And why?
- You have almost [X amount of] followers. Do you know who these people are? Who do you imagine them to be?
- When you tweet, who do you think is listening? Do you keep any particular people in mind when composing tweets?
- How and why do you think you became such a hit on Twitter?
- Which of your tweets have received the most reaction, good or bad?
- Have reactions to your tweets prompted you to change your behaviour on Twitter?
- What do you think makes the South African Twitter landscape unique?
- Do well-known people such as yourself face any specific challenges on platforms such as Twitter? And what are the advantages?
- Do you think South Africans are more cautious to tweet about things such as politics and race and that they are careful not to make prejudices apparent, given our country’s history?
- Is it possible for you to separate the professional from the personal on Twitter (if you even want to)? If yes, how?
3. Questions on Twitter
I created a Twitter account for the project called @TweetMzansi so that I could ask people questions on the very platform that I was investigating. Questions were similar to the questions asked in the questionnaire. Additional questions such as “Do you care if people unfollow you” were added as the research progressed. Very few responses were received in this manner.
Tweeting links to the survey proved to be more successful and many retweets were received. I believe that this contributed to many of the survey responses from respondents outside of my network.
The follower count for this account is growing daily and hopefully it will be used in the future to create a conversation around Twitter in South Africa.
4. An ethnography of Twitter
Lastly, I also relied on my own experience and observations using Twitter. Virtual ethnography is ethnographic research that is carried out in the online setting. I have been studying the online community and culture of Twitter for quite some time. I joined Twitter in 2008. I have 1,434 followers, have made more than 9,000 tweets, and follow 899 people (of which most are South Africans).
Although this didn’t have a big effect on my research results, it did make me more confident about the validity of my findings as many of them I have experienced firsthand.
2. Creating a brand
Twitter is a changing platform and with changes in the network come changes in people’s behaviour. From the inception of this project I intended it to be an ongoing study and discussion. I thought this might be easier accomplished by creating a brand. And so Tweet Mzansi was born. I created a website (www.tweetmzansi.com) and twitter handle (@TweetMzansi) for this purpose.
Another motivation for creating a brand with a website and a Twitter handle, was that it would enable me to better market my research and hopefully receive more responses to my questionnaire.
3. Marketing techniques
In order to promote my research and survey, I wrote a few articles about Twitter and Twitter statistics, which I placed on the website. I used the @TweetMzansi account to tweet these articles, as well as other interesting facts and links.
I transcribed all my interviews and converted them to articles for the website. At the bottom of each article was an invitation to South Africans to add their voice to my research with a link to the survey. This was another reason for choosing to interview popular Twitter users. I tweeted the links to interviews with them along with their Twitter handles and a link to the survey. These tweets were retweeted many times and lead to many survey contributions.
I will also use the website and Twitter account to publish and promote my results and will also pitch my results to various publications and websites in South Africa for publication. Hopefully this will contribute to keeping the conversation going.
When studying large populations of people, quantitative approaches dominate and samples are most often stratified based on demographic information — usually census data. When it comes to studying Twitter, it is impossible to stratify data based on demographic information, as Twitter does not provide this. This introduces the difficult question of how to negotiate between the practical logistical limits that most research projects are subject to in order to infer from the sample to a larger population (Rieder 2012).
There is a difference between probability samples – where the (positive) inclusion probabilities for all units of the target population are known in advance – and non-probability samples. Probability samples in Internet surveys are highly affected by the problem of non-coverage and sampling frame problems. The first problem arises from the fact that not all members of the general population have access to the Internet (Fielding, Lee and Blank 2008). As this study was only concerned with studying a subset of Twitter users (South Africans) for which there is no complete demographic information, the sample is a non-probability sample.
I did endeavour to include responses from a variety of South African Twitter users in terms of demographics, behaviour and follower counts. This was one of the biggest obstacles that I had to overcome in my research. To my disappointment I discovered that my Twitter network was not as diverse as I had hoped. Most of the responses were from white South Africans. I really wanted the responses to be more representative of the various cultures and groups within South Africa. In order to try and overcome this problem, I asked a very popular black Twitter user Khaya Dlanga (@khayadlanga) if he would promote the link to my survey. Dlanga has almost 100,000 followers and his tweet assisted in drawing responses from a diverse selection of South Africans.
Another problem was the distribution of the questionnaire. According to Fielding et al. (2008), invitations to Internet surveys are most conveniently distributed using e-mail. This causes severe frame problems because there are no e-mail directories of general populations of Internet users (and in this case Twitter users) that might be used as a sampling frame. Both coverage and frame problems can significantly impact data quality and should be adequately reported when disseminating the results of research (Fielding et al. 2008). However, Fielding et al. note that the problem of inference from non-probability samples should be considered as a purely statistical issue, which also applies to other survey modes.
Fielding et al. 2008 also note that sharing a link to a survey might lead to selection bias, which is out of the researcher’s control, and that a measurement error can arise in Internet surveys because of questionnaire design, but also because of the respondents or the survey mode itself. Respondents’ motivation, computer literacy, abilities, privacy concerns, and many other factors may influence answers. Privacy concerns were mostly dealt with by giving users the option to submit answers anonymously. The survey didn’t elicit any sensitive information such as contact details. The only identifier that was used was a respondent’s Twitter handle, and this was also voluntary.
Lastly, researchers must remember that there is a difference between reported behaviour and actual behaviour. The value of data collected depends on how truthful respondents answer questions. Respondents sometimes feel pressurized to respond in certain ways. However, my ethnographic fieldwork on Twitter leaves me fairly confident that responses received do reflect reality. Respondents also had the option of submitting survey responses anonymously.
5. Measure of success
In the context of my research project, I endeavoured to achieve the following:
- I wanted to create a sample of South African Twitter users that was representative of all cultures and people in South Africa.
- I wanted to gather enough data and insights to enable me to make some assumptions and generalizations about the behaviour of South African Twitter users.
After cleaning the data (removing duplicate, invalid and otherwise faulty entries) 217 valid survey responses remained. Because my initial target was 100 responses, I am very pleased with this number. In the absence of clear demographic information about Twitter users and Internet users in South Africa, it is impossible to say with certainty how representative a sample is. However, I am very satisfied with the distribution of the sample and I am pleased that it includes respondents from various ages, genders, races, languages and occupations.
Of the 217 people who completed the questionnaire, 124 people requested to submit their answers anonymously and 93 people were happy for their Twitter handles to be used in the research results.
Responses were rich and detailed. I believe that the combination of surveys and in-depth interviews were effective in providing higher validity and explanatory power of the collected data. The responses were also supported by my ethnographic fieldwork. I am confident that my data allows for at least some generalizations about the Twitter behaviour of South Africans.
 The only limitation is the number of cells that can be used for the storage of results, as a Google Document is limited to 2,000 cells. This can be overcome by exporting and deleting data to free up more cells, or by creating a new destination for responses.
 The survey can be seen in the Appendices.
 Unique identifier or name on Twitter. E.g. @sarietha
 A private message that can only be seen by the intended person. DMs can only be sent to people who are following you.
 Also known as OpenRefine.
 See Appendices for examples.
 See Appendices for this tweet.
 Demographic information about respondents is available on request.