[EM] Ballot design (new simple legal strategy to get IRV)

[Kristofer Munsterhjelm]

On 10/10/2015 11:43 PM, robert bristow-johnson wrote:
> On 10/10/15 5:30 PM, Juho Laatu wrote:
>>> On 11 Oct 2015, at 00:10, robert
>>> bristow-johnson<rbj at audioimagination.com>  wrote:
>>>
>>> On 10/9/15 4:00 AM, Juho Laatu wrote:
>>>> I just note that there can be also simple ballots like in
>>>> https://en.wikipedia.org/wiki/Open_list (see the picture). If one
>>>> wants to expand that to ranked ballots with high number of
>>>> candidates, one could simply allow the voter to write multiple
>>>> numbers in the ballot instead of only one. One could thus cast a
>>>> ranked vote by writing few numbers, e.g. "23 74 74 5 234 321".
>>>>
>>> and a machine is gonna scan that?
>> Manual scan by default. In Finland the open list ballots (bullet
>> votes) are counted today by hand right after the polling station
>> closes (within say 1 hour). I'd propose to do the same also with the
>> (extended) ranked ballots. With ranked ballots you need to introduce
>> also computers to store the ranked data, and the process takes a bit
>> longer.
>>
>> Machine scanning could be used too if it is reliable enough. Maybe so
>> that machines use human help for ballots whose interpretation is not
>> obvious.
>>
> 
> listen, here in Vermont, i have worked on a few different election
> recounts where we had to examine thousands of ballots by hand.  many
> people mark even bubble or bullet ballots so poorly that even that fails
> machine scanning.  i know there is Optical Character Recognition (OCR),
> but i would not trust that to recognize numbers written by voters on
> ballots.  numbers like "3" and "8" get confused.  and other pairs, like
> "1" and "7" and "5" and "8".  or "5" and "6".

If you have check digits or checksums, that gets rid of most silent
failures; how many they get rid of depends on what kind of errors you
want to correct and how much redundancy you're willing to add to the ID
- e.g using five-digit codes for ten candidates can handle more errors
than using four-digit ones.

The machine could then signal an error if the ID doesn't read as valid
to its software, and then have a manual counter (or multiple) verify the
ID by hand.

I don't know how good handwriting recognition is at the moment, though;
perhaps it would have to be all manual. But there are at least ways to
avoid silent errors like one candidate being mistaken for another. I
also imagine one could not use digits that look too alike, for instance
using only the digits {1, 2, 4, 6, 8, 9}.