Voting inaccuracies in NA 256 and NA 258: We need a better system!

[W.11]

Voting inaccuracies in NA 256 and NA 258: We need a better system!

AFIS and OCR are technologies that are not meant for bulk vote verification. Surely there are people in the election commission who were aware of the limitations of the software before they decided to go ahead and deploy it anyway. PHOTO: AFP/FILE


The May 11, 2013elections were historic for Pakistan. The eyes of the world as well as the hopes of the country were pinned on them. These elections deserved the full attention of officials who were responsible for their execution. However, the fact that inaccurate software was used is inexcusable.
All software used to decide the fate of a country should be state-of–the-art and the best possible available, not technology that has been discarded as being inaccurate, for what is the point of even having elections if the date is not correct?

On October 7, 2013, the National Database Registration Authority (NADRA) submitted its report of thumbprint verification for NA-256 (Karachi) to the election tribunal. Few weeks ago, Nadra also submitted a similar report for NA-258 (Karachi). Data on both reports showed substantial irregularities in both constituencies during the May 11 general elections.

The data for both constituencies (NA-256 and NA-258) looked distinct at first glance but, when subjected to a thorough analysis, the similarities became obvious.

The results from NA-256 and NA-258 are shown in the Table.

Initially, Nadra was capable of matching only 100 fingerprints per day but that capacity has been upgraded to 100,000 fingerprints per day at the cost of Rs10 per vote. According to the Chairman of Nadra, Muhammad Tariq Malik,

“Increase capacity has been achieved through Automated Fingerprint Identification System (AFIS) software development by Nadra’s own engineers.”​

This prompted me to research AFIS further.

Automated Fingerprint Identification System (AFIS)

The FBI initiated the project for automating fingerprint identification process in the United States in early 1960. In 1975, the first AFIS software was developed. In 1999, FBI upgraded their system from AFIS to second generation Integrated Automated Fingerprint Identification System (IAFIS) which is largest finger printing database in the world. In 2009, the FBI announced plans to replace IAFIS with a Next Generation Identification System (NGIS).

According to the FBI, 98% of all IAFIS transactions required a manual review and approximately 15% of all NGIS transactions still require a manual review.

AFIS technology
AFIS is an open-source software SDK (Software Development Kit). A few years back, AFIS technology was proposed to the Yemeni government for voter verification but the project did not hit the shelves because of high rate of inaccuracy and discrepancies. Without manual review the inaccuracy rate was 90%, only 10% of the records were truly matched. Later AFIS replaced it with the second generation IAFIS. The Nadra report data shows discrepancies of 7.5% and 8.0% of total verified votes in NA-256 and NA-258, respectively, which is under the accuracy rate of automated AFIS technology without manual review.

Optical Character Recognition (OCR) Technology
OCR is the mechanical or electronic conversion of scanned images of handwritten, typewritten or printed text into machine-encoded text. This technology is used by Nadra to convert and extract handwritten NIC numbers from vote counterfoils and run them against the Nadra NIC database for verification.
According to leading OCR software companies, the error rate for handwritten documents is between 15%-20% depending on the image quality and writing style. Nadra reported data that shows 14.2% and 13.3% of fake ID numbers in NA-256 and NA-258, respectively, which is within the percentage of software error rate.

For AFIS, the standard scan image quality is between 500 to 1,000 dpi, or dots per inch, for a better and desirable result. But currently, NADRA is scanning vote counterfoils images at 400 dpi, which definitely increases the capacity rate and, incidentally, the error rate.

Most importantly, Nadra civilian AFIS uses flat fingerprint, in which the finger is pressed down on a flat surface but not rolled. All the vote counterfoils were verified against the flat fingerprints. Any roll fingerprints/poor quality fingerprints/low ink fingerprints on counterfoils were categorised as “bad quality” fingerprints or “failed authentication”. The Nadra report data shows 71.2% and 70.3% “bad quality” fingerprints or “failed authentication” in NA-256 and NA-258, respectively.

AFIS and OCR are technologies that are not meant for bulk vote verification. Surely there are people in the election commission who were aware of the limitations of the software before they decided to go ahead and deploy it anyway.
More importantly, the people should have been apprised of the possibility of these discrepancies ahead of time to avoid confusion after the results were in. Election is a process which demands transparency for validity. In the current volatile political atmosphere of Pakistan, deciding to use inaccurate and obsolete software proves that the citizens of Pakistan are still not the factor in these important choices.