Polling on Twitter

There was an article published in the Business Day titled Diversity makes it tough to take the nation’s pulse. After reading, I thought, surely there must be a way?

We’d recently moved over to Google Apps for work. So I was keen to try out Google Forms and Sheets to see what could be done. So began the experiment.

Even though twitter has introduced a polling feature, I’m always hesitant to read into the results. Diversity makes it tough to take the nation’s pulses. The lack of any sort of demographic information makes it incredibly difficult to know if the sample, no matter how large, was accurate.

So what would happen if one put together a form that collects demographic information and then poll whatever? One would hope that would lead for a far more accurate sample, less dependent on size (maybe?)

Surely if one could collect enough responses, then it could be possible to find some appropriate weighting of the different demographic categories to accurately represent the national pulse – despite having a biased sample?

There is going to be a Local Government Election in August, so now seemed the perfect time to poll for political party support. We could use these responses, and check them against actual results. Yes,  not everyone votes in LGE, but it’s better than nothing. So long as the results collected were in the same ballpark, I would see it as a success.

I wasn’t expecting to get thousands of responses – though I desperately hoped I would. Nor was I expecting a perfectly representative set of responses – though I never expected that the responses till now would be as biased as they were.

So far 65 people have filled in the form. Of those 65 people,  59 were white. My form, that I had hoped was “How South Africa Votes”, has turned into “How White People Vote”.

Here is a table of How White People Vote, so far,

Party Support (percent +/- error)
ANC 7.8 +/- 7.6
DA 61.9 +/- 12.8
EFF 6.6 +/- 6.6
IFP 4.3 +/- 4.3
NFP 4.3 +/- 4.3
UDM 4.3 +/- 4.3
Other 5.5 +/- 5.5
None 26.9 +/- 11.8

This may surprise some, but the majority of white people vote DA, and the next biggest segment is None. So it’s DA or nothing. The odd results for IFP, NFP, and UDM are simple because no one selected them. ‘Other’ has only one response.

In terms of response rates, 123 people clicked the link to the form, but only 65 answered. Which is rather good response rate. But the form has no where near enough diversity to display anything truly meaningful.

There are all sorts of problems with the form, that I’m not going to get into. This was really a proof of concept. What would be great to get out of the data, something a lot of people suspect, but has really not been confirmed, is what is the greatest predictor of party support?

Gender should be useless. But what of race, home language, province or age? What if the greatest predictor is home language and age? Where the old of a certain language vote X, but the young of the same vote Y? This would be exciting to know!

The form is still open, and I’m going to try using twitters sponsored tweets function to get more responses. We’ve never really had access to such a data set before.

I’m going to give it a few more weeks, and then release the data set. Don’t worry, it is completely anonymous. Not that anyone has actually asked to see the data. Maybe I’m the only one excited to see it? Anyways, it will exist, and those that are interested can contact me.

I’m sticking to twitter as a medium. Trying to milk as much meta-data out of the form as possible. If, despite best attempts, one fails to collect accurate information, could we assume that there are limits to twitter being the pulse of the nation, and thus limits to twitter polls?

If you have yet to answer the questions, please may you. It doesn’t matter if you’re white. If you’d like to help gather more responses, please use this link: https://goo.gl/forms/VcXeKzaQ9q, otherwise the form is embedded below:

On Google Forms. The use of the date question type confused some people, mostly because it has the American date format of month/day/year, and on some mobile devices it just didn’t work. So try avoid that if you plan on using Forms. Also, there were some issues with the drop down selections showing on mobile, but not functioning. But in terms of data collection, Google Forms is super simple. As the responses are made, they are written straight to a sheet, and you can set up your analysis to automatically update as new results come in.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s