*If Edwinn Starr had sung this lyric about SNOW instead of WAR he would have been wrong (he also wouldn’t have made such a valid political point…but I digress). Much as we moan about it snow can be an invaluable way to track down local social media users!
When it snows in the UK, the whole country grinds to a halt.
Popular opinion has it that this is because we don’t get enough snow annually to justify the expense of investing further in our snow handling infrastructure. However I have another theory…perhaps it’s
because the whole country stops what it is doing and starts talking about the snow…!
We do love a good chat about the weather – and this can be just the tool to help you identify local social media users (I’ll concentrate on twitter, but the same principles apply all over social media land). This blog is basically going to tell you that you will find good social media contacts if you search for location names when it is snowing…but bear with me and I’ll prove it to you! (and you’ll get an extra snow-related tip that is worth the read – I promise!).
What’s the story?
Since 1 December 2012, I have had a constant search running on Twitter for 6 Surrey placenames: Guildford, Camberley, Epsom, Woking, Redhill, Staines, Reigate. I used a fantastic (paid for) service called Tweetreach that will just collect any tweet it can find that matches a particular set of search terms. This data collection was as part of a work project to spot patterns in tweet volume which I’ll try and blog about that some other time.
I downloaded the tweets and did my analysis in Excel. Tweetreach will give you huge amounts of useful information, but for my purposes I was looking at the number of tweets per day. Between 1 December and 20 Jan there were 174640 tweets (including RTs) matching one of those search words. It amazed me how much the daily count fluctuated – between 1667 (Christmas day – guess people have better things to do) and 5385 tweets per day. I graphed the daily volume (I whipped out RTs because they are less likely to be by a person in that location).
There were two huge peaks – 5 December and 18 January. Both days when it snowed, and there were lots of mentions of snow (the blue area is mentions of snow in my dataset). SO the tip to “search for placenames to find local tweeters” works far better if it is snowing.
BUT we can go further thanks to the wonderful Mr Ben Marsh
Ben is the gentleman who created the #UKSnow Map. If you are not familiar (and really you should be!) this is where people tweet the first part of their postcode, the hashtag #UKSnow and a rating out of 10 for how heavy the snow is. SO, you can also find local tweeters by searching for #UKSnow and the postcodes that are of interest to you. Of course that only works when it is snowing…
There is a bit of a dilemma here about whether this is “prying” too much. My take is that, if you follow your local target (promptly after their tweet), and actually engage with them through social media (rather than just sending them lots of spammy messages), this is OK as a tactic. I’d also suggest automating the following/interaction process is a no-no. I guess you need to consult your own moral compass about this whole tactic though!
The #UKSnow search also neatly sidesteps some of the “false positives” problems.
When you actually read some (not all – I am not that sad!) of the tweets in my collection, you spot some patterns – and not all of them are useful to me…
Epsom is a lovely town, which has also given its name to Magnesium Sulphate aka Epsom Salts! It is astounding how many people talk about Epsom Salts on Twitter – There were 35955 mentions of Epsom in my sample, and 42% of those also included “Salt”. Luckily this is easily cleansed out in the search (“Epsom -Salt”) – but I didn’t check my data early enough to make removing it a viable option. Instead I had to code around them in Excel.
Woking is a common misspelling of “Working”, and some people misspell “Stains” as “Staines”. There isn’t a lot you can do about that, other than assume that the rate of misspelling is constant (and fairly low), and assume they cancel each other out.
Disclosure: I will not be following (either professionally or personally) any of the people I have captured with my twitter search, as a result of their being in the search. Because I feel that is spammy, and there are 87576 separate individuals (which is waaay too many to follow anyway!)
The next thing I want to try with my data is to identify how many people tweet in each of my target towns, using “capture, mark recapture” methodology (which I used loads to count field voles during my biology degree). Unfortunately the maths for correcting the data (as not everybody tweets with the same frequency) looks a bit complex for me!