Monday, May 10, 2010

PCOS is not the bottleneck: Additional Data on Voting Throughput of Clusters (Election 2010)

Did the Comelec commit a huge omission on election day?

Sure, the presidential elections is a huge political process, a huge exercise in patriotism, and a game changing decision making process whose results will affect not only our generation but also the generations to come.

But isn't the elections a huge manufacturing process as well? It is a customer interface of great magnitude. The voters are being processed by a system that is monopolistic, they have no choice on the matter.

Unfortunately, it seems nobody has ever conducted a study of the voting process from the point of view of operations management.

This is a huge omission considering that this process involves 50 million people undergoing a poorly studied and poorly managed physical process.

Comelec and the management groups missed out on designing and conducting a time and motion study of the actual voting process on election day.

Queuing theory, line balancing, critical paths, load distribution, capacity utilization, probability density functions - this could have been a huge management study.

For example, like what I have done below and in the previous blog, the PCOS count and the time elapsed could have been taken down every 30 minutes by the youth volunteers and the by the BEI themselves by filling out a simple form. This process takes less than 10 seconds, every 30 minutes.


All those data points can actually be used to do curve-fitting. When is the peak time? Where is the bottleneck? Is the process quadratic? Exponential?

Can you imagine if Comelec and the management associations had this information collated on the 76,500 PCOS clusters? What a wealth of management information!


In a very simple way, I attempted to do this by gathering actual data from the BEI's (had some extra help, of course) and the survey below reflects my feeble attempt at extracting scientific data and illustrating the kind of studies that can actuall be done.

But then there's always 2013. I can volunteer to help Comelec design such a study. I am not an expert but at least I have already done here what the Comelec and other agencies and private management associations could and should have done. Although, this study below has obvious limitations.

Is the voting process a Poisson process or a Gaussian process? How do we manage the queue? Should we encourage the voters to come early and frontload the process thereby having a peak load at 8am with almost no load at 5pm?

The PCOS is not the bottleneck of the voting process. In fact it has been idle most of the time. The reason for the low throughput is the slow passing of the filled-up ballots by the voters. This is due to the low number of voters being processed at one time.

Offhand, one of my suggestions is: Instead of 10 voters at a time, Comelec should allow 30 (or 15) voters at a time (find the space). After one goes out, one voter must be allowed in. In this manner, at any given minute somebody is finished filling up the ballots and feed it to the PCOS machine.

****************************************************************


Note:
1. This report should be read in conjunction with the previous blog on voting time.
2. Voter turnout in this school (19 clusters) is only 67.3% (in highly urban district) where the next elementary school is just about 1 to 2 km away. (12,332/18,329 - this is not a small school)
3. Did the voters leave but did not return?
4. The lowest turnout is 57.4% (560/975)
5. The actual throughput of the entire school of 19 cluster is 1.06 min/voter (17.47 hours for 1,000 voters)
6. As of 6:40 practically no more voters.
7. PCOS machines is not primary reason for low throughput. It is the low number of voters passing their finished ballots (instead of 10 voters at one time, they should make it 30)






Survey Time from 6:39PM to 6:54 pm (19 clusters of Mabolo Elem. School Cebu City









ClusterTime TakenTime Elapsed (Minutes)PCOS Count Read (Ballots)RegisteredTurnout PercentageMinutes per VoterFinal Results?
1576:3969970698571.7%0.990085Final
1586:3969970497971.9%0.992898Final
1596:4170172798374.0%0.964237Final
1606:4170172099472.4%0.973611Final
1616:4270269899470.2%1.005731Final
1626:4370373799374.2%0.953867Final
1636:4470470199070.8%1.00428Final
1646:4570571199171.7%0.991561waiting*Fastest
1656:4670665898566.8%1.072948Final
1666:4670667098767.9%1.053731waiting*
1676:4770770199570.5%1.008559waiting*
1686:4870865798766.6%1.077626Final
1696:4970968999569.2%1.029028Final
1706:5071058994062.7%1.205433waiting*Slowest
1716:5171156097557.4%1.269643Final
1726:5271261299261.7%1.163399Final
1736:5771757098857.7%1.257895Final
1746:5371345378657.6%1.573951Final
1756:5471446979059.4%1.522388Final









* waiting less than 5 voters as of survey times












Note: The minutes per voter not reliable bec no more voters at survey time

The minutes per voter for those with waiting are applicable











Average Minutes per Voter (for clusters with waiting voters):1.06482117











Average Turnout
   12,332.00    18,329.00 67.3%