Smart SMS Parsing for Better Election Observation

Kenya's election is over and was largely peaceful, even as there are ongoing court challenges. We @NDITech assisted the Kenyan civil society organization, ELOG, in it's election observation effort on Election day so had an inside view of this much-anticipated and closely-watched election. NDI specifically supported ELOG's data collection effort where observers gathered process and incident data at polling stations around the country as well as vote share data to verify the results publicized by Kenya's electoral commission, IEBC.  As the IEBC found out the hard way, it’s not easy to collect electronic data from tens of thousands of polling stations around the country.  ELOG’s observers were trained by master trainers to collect relevant data and then send coded text messages for processing to a central database. 

This process requires an intelligent parsing engine with helpful auto-responses back to observers.  In the image displayed, the observer sent at 13:26:47 local time, “D0Q0” and the response from the system was “Thanks for your message. It was not recognized as a valid report. Please review, or contact supervisor or ELOG center”.  As you can see through the displayed sequence of responses, the observer learned quickly and was able to send in complete data for all questions in a section, while correcting missed and incorrect responses for Questions D&Q.

On election day, more than 7,000 ELOG observers were present at polling stations randomly selected thoughout the country and they sent more than 30,000 text messages.  Self-correction and auto-response behavior significantly reduced the amount of work required for cleaning data and correcting for user errors. While it's still necessary to have around-the-clock shifts of data clerks who work the phones and call observers when issues arise, increasingly sophisticated feedback systems can reduce the overall organizational load.

Image courtesy: Jared Ford.