You are here: Home / Pakistani Politics / Fingerprint Verification – Part 3
Fingerprint Verification – Part 3

Fingerprint Verification – Part 3

This is the third in a series of posts on the ‘fingeprint verification’ of ballots in Pakistan. I first wrote about it in September 2013, when the whole exercise was picking up steam as some forensic way of settling electoral fraud debates. The second part was in July 2014, and expanded on the misrepresentation of facts from the NADRA reports on fingerprint verification of ballots. This will summarize earlier points and add more details to some. If you’re in a hurry or this is too long, skip to the fifth or seventh para.

First, fingerprints are not unique. They are highly unlikely to be common amongst unrelated people (families often share similar traits) but fingerprint uniqueness is not absolute scientific certainty. Just like there are identical snow flakes, there are highly similar fingerprints. It is a probabilistic exercise and automated systems return ‘matches’ based on that probability (unlike DNA, which is unique). AFIS systems for criminal investigations return a set of possible matches from which an investigator testifies to the matchness of one after examining them.

Second, AFIS systems (automated fingerprint matching) have limitations based on a milieu of factors, including the resolution of the original print, the source of the print to match, the resolution of the scanned print, the exposure it went through, etc. Accuracy of latent prints can be upto 99.8%, but even more near certainty is achieved if say all 10 fingers are verified, and criminal cases usually have 2-3 prints pulled from objects. What the high numbers hide are legitimate cases where a match was rejected by the AFIS due to a high threshold kept to reduce false matches.

Third, false negatives are very common. The four types of cases that can come up during fingerprint matching are: true positive, false positive, false negative and true negative. The first one means a correct match was done, the second one means a false match was done i.e. person X’s fingerprint was matched to person Y’s, the third case is where a fingerprint from person X was rejected as not belonging to person X, and the last one is where person X’s fingerprint was correctly rejected as not belonging to person Y.

In one study, where the sources of the print being matched to an original was different, 2.9% cases were false positives. That is, 2.9% of the fingerprints were incorrectly matched to another person. 10.5% of cases could not be matched and deemed inconclusive. Even more worrying, there were 7.7% false negatives i.e. the same person’s fingerprint could not be matched to an original. The study had 12 trials and in 2 of those batches, more than 50% cases were inconclusive. They have a high repeatability issue too i.e. the same evidence presented again can lead to a different conclusion making the exercise inconclusive.

Another study showed false negatives to exist at the same rate, 7.5%. And we’re talking experts using candidate exemplars from AFIS and suspects to match against latent prints here, which is more accurate and less prone to error than just AFIS returning a single match (FBI’s AFIS returns a 8 candidate pool from which an examiners chooses one).

Fourth, we know these things about fingerprint matching because the technique has long been in use in criminal trials and courts tend to set a high bar for evidence. Studies conducted and details of technical parameters from LEAs like the FBI have given us this data. From NADRA, we know nothing. They apparently use AFIS to generate a single match, not a pool, and thus are more prone to errors of all kinds: false positives, false negatives and exclusions.

NADRA had claimed to have ‘ingenuously’ upgraded ‘software’ to scan & match 100,000 fingerprints, up from 100 a day. Earlier, they were aiming for 500,000 a day. As I’ve written before, NADRA is a PR machine, the first and foremost goal is to achieve good press and fool a susceptible populace with gizmo magic. See how NADRA claims to have ingeniously developed ‘software’ without any financial input from GoP? Well, Rs 90 million was released for it. NADRA had done this exercise earlier for the Watan Card (Citizen Damage Compensation Program) but required a suspicious amount for a software upgrade that no one knows the details about.

I had mentioned these cases before, where NADRA admits that it does not have the “biometric record” of around 4 million females. They have either issued IDs without taking pictures/fingerprints of simply lost the record. NADRA accepted that “relaxation had been provided to women against provisioning of their photographs considering social values.” How were these women placed on electoral rolls then? (Pictures were printed on rolls for verification at polling stations). For re-issuance of a CNIC, NADRA runs an identity check and since the CNICs are new and few people have had to get new ones made after the old one expired, yet we know cases (view one online) where the same person’s identity could not be matched again. In that case, the source of the fingerprints (high resolution optical scanner) and the number of fingerprints recorded is really high, yet identity could not be verified.

Fifth, for an exercise as political and problematic as electoral audit, one would expect a pilot exercise to have been done to establish a baseline for various errors and limitations. Take one, or more appropriately four to six, polling stations and run them under ordinary working conditions of 800 to 1,200 voters in eight hours with the same amount of administrative staff. Identities of voters being known in this exercise, statistically relevant error rates of both false negatives and false positives would show up. We would come to know unreliability of a single thumbprint on a paper, done in haste, as a electoral audit mechanism. But this was never done, and NADRA took in the money it wanted from the ECP (via the candidates).

Sixth, AFIS systems as stated previously, also have false positive (Type I) and false negative errors (Type II). Source of print to be matched, rotation of the print and the visibility/contrast of the core of the print have high correlation to the ability to match too. In real world conditions, a single thumbprint deployed by an ink press on a piece of paper is less likely to return matches than if recorded using a high resolution optical scanner (and even that can and does lead to errors, as anyone having used a fingerprint based office time entry system knows). In NADRA’s case, we do not know any technical details. We do not know what thresholds they maintain (lately, they claimed that they use 15 minutiae), what are the technical parameters of the scanning or even if how the ballots are being scanned (optical scanner automated digitization of the CNIC on the ballot or a human verifies/corrects/enters the CNIC from the ballot after thumbprint is scanned?). Are the acceptance/rejections thresholds being kept at the same level between various constituency verifications or even in one day, being changed day to day at whim, so as at to produce viable and press positive reports?

NADRA’s exercise is completely unsupervised and unaudited. For a first time exercise, this is irresponsible and dangerous. It is in NADRA’s interest to show low match percentages, so as to get good PR (it’s primary aim) and not seem like an extension of the government but some supreme technocratic dream.

Seventh, the curious case of the ‘magnetic ink’. While parties aiming to delegitimize the electoral exercise assert that the ‘magnetic ink’ was not used, the ECP asserts that it was infact widely used. So where did the issue of not using ‘magnetic ink’ come from? NADRA says it wasn’t as widely used as POs reverted to using ordinary ink pads in many cases (ran out of ‘magnetic ink’, fewer supplied than required etc?) but it is in NADRA’s interest to claim this when it is unable to match large number of thumbprints. NADRA keeps finding the same press release for re-release as well. See this statement from Chariman NADRA, Tariq Malik (more on him later) for example, “It is not just Nadra’s software but any software in the world cannot authenticate such fingerprints unless the specified ink is used, he explained”. Only in Pakistan can someone get away with making laughable statements like that, and without any counter-questions.

Now coming to the curious case of what the hell is the ‘magnetic ink’? Supposedly, it would have had magnetized particles in it, and that would increase the readability of the fingerprints and lead to less smudging. I have been unable to come across this notion anywhere else except the case of Pakistan GE 2013. Magnetized ink is used in OCR sometimes, like banking cheques (scanning and verifying purposes) and magnetized powder is used for latent fingerprint pulling purposes (it avoids direct contact with the surface and thus preventing damage to the print) but could not find anywhere else, this notion of an ink with magnetized particles being used to enhance fingerprints. MICR is used to scan magnetic ink or magnetized particles and NADRA does not have that, but rather OCR an optical scanning only. Magnetic ink should have had no relevance at all to the fingerprinting exercise. PCSIR, ECP and NADRA supposedly say that the use of magnetic ink was not even necessary. Who is one to believe? Their previous statements or the new ones? Was it just a scam to spread the money around or did NADRA really develop insight that use of a specific ink would increase ability to match fingerprints? If they did, they should let us know the details.

Eighth, the real world conditions of the elections are not ideal. Rapid processing by humans at polling stations leads to errors and many procedural violations occur. People are being hurried through as they press their thumb down on an ink pad and then on a paper. There is very little care for the orientation of the print or how big a part of it is actually delivered onto the paper, which does have real correlation to the ability to match. Smudges develop and render the print near useless. The ballots are stuffed one on top of another, packed and then re-opened (ink should have dried by then), counted and then packed into bags instead of boxes this time and shipped off to the ECP for safe keeping. After months, and in some cases more than an year, the ballots are re-opened and scanned. They are obviously not being kept in humidity controlled environments like some pristine archival material. Humidity, temperature and the conditions take a toll on the paper, and the thumbprint on it too. This keeps reducing the number of prints that could be matched.

CNICs are being recorded onto the ballot paper by humans, and there are likely human errors in copying a 13 digit number too. If NADRA is OCR-ing that or even if humans are recording it, there are possible errors in the CNIC number being further introduced at the verification stage too. Say a legitimate person, called A, came and voted legitimately and either the polling station staff recorded one digit wrongly onto the ballot paper or later NADRA erroneously recorded the CNIC when doing verification. That person’s vote should not directly be deemed fake and all efforts should be made to see if it was a legitimate one. Similarly, if that person’s fingerprint is not matched at a later stage by NADRA, his/her vote should not be deemed fake or bogus.

Ninth, the ‘massive rigging found’ is nowhere near massive. In NA-118, 8.76% of the votes had wrong CNICs on the ballots (see above for error at two stages). NADRA did not run the prints on them against it’s entire database to see if a near-matching CNIC number would return a match, thus suggesting that a human error was made. Another 0.3% were deemed to be from CNICs not in the same constituency. Were these because a wrong CNIC was recorded and it matched a real one and thus deemed to have been voted in the wrong constituency or a genuine stuffed ballot in a different constituency? A further 0.58% ballots had no fingerprints. 0.095% were by unregistered voters (or another CNIC error making it seem like an unregistered person, though less likely). 55% of the ballots could not be verified by NADRA for what it says is bad quality prints, but as explained earlier, inconclusiveness exists very commonly and NADRA is possibly and probably adding those to this list for PR purposes. Only 0.3% were found to be multiple balloting (false positives are rare, so we can ignore them).

Definite ‘Bogus’ or ‘fake’ votes = Multiple voting + ballots without CNICs + counterfoils without fingerprints + CNICs not registered in the constituency = 462 votes out of 46,118 = 0.9%

Probable ‘Bogus’ or ‘fake’ votes = CNICs not registered in the constituency = 142 out of 46,118 = 0.3%

Percentage with possible fake/bogus votes = Wrong CNICs = 4,043  out of 46,118 = 8.76%

Percentage with accurate CNICs and fingerprints on counterfoil = 41.471 out of 46,118 = 89.92%

I took a constituency where PTI is the victim party, and in Punjab, because that’s where the real trouble is at. I do not question that elections in Karachi did not have high ballot stuffing or voter suppression and manipulation did not happen in Balochistan.

I did not mention irregularities like votes missing from the ballot boxes (suggesting count manipulation, more on that later) but it doesn’t help if you produce a document based on your sources (official not released yet) and add ‘bogus votes found and rigging established’ at the bottom to what you want to make look like an official document. Or when you include bad quality fingerprints in bogus (again, official document not released yet, as case still in tribunal). Moreover, in specific cases, specific polling stations were called out, not a random collection, and extrapolation to the entire constituency is wrong.  Lumping all cases together is wrong.

Imran keeps repeating this ‘60,000-70,000′ fake/bogus/unverifiable votes’ thing (wording changes daily) and questions if an election should be deemed fair if such a large number cannot be verified. M

As stated earlier, studies show existence of 7.5-7.8% false negatives and upto 13% inconclusive matches. If one were top repeat this exercise of on average NA constituency with 200,000+ polled ballots, there would be at least 15,000+ votes where a legitimate person voted, all data on the ballot was accurate yet a match could not be returned. A further 26,000 prints would not return a match due to technical limitations and minimum requirements. Heck, a 0.1% false positive rate means 200 fingerprints would wrongly be identified as belonging to someone else‘s i.e person X matched as person Y. This can be upto 2.9% ins some cases and with NADRA’s one match per AFIS, definitely on the higher side. A thumbprint kept on a paper over time would have even more legitimate ballots being deemed bogus by those whose sole aim is to delegitimze the elections.

Tenthly, the vast majority of elections have not been called into question by the candidates. See the previous post for tribunal numbers as of July. For most of the cases in Punjab, the winning margin have been enough enough so as to have been highly unlikely a case of fraud. Go through this, to see how significant a number are cases where the winning margins (and zabt zamaanat of losing candidates) suggests fraud could not have played a major role in electoral victories. Even cases of winning margins being lower than rejected vote counts shows no evidence of widespread fraud in favour of the PML-N. And since Imran has talked so much about the winning margin > rejected margin as proof of fraud, and said rejected votes had increase enormously, it would be relevant to know that the number of rejected votes as a percentage of total polled was the lowest in Punjab, at 2.94%, 3.2% in KPK, 3.7% in Sindh and 5.5% in Balochistan. The increase in the number of rejected votes b/w 2008 and 2013 was high, due to higher turnout. The percentage however was the same, it actually reduced a little bit from 2.97% the last time.

Lastly, Confirmation of voter identity from a ballot is not as supreme in electoral audit as it might seem. This is because ballot stuffing is difficult to do at a very large scale and count manipulation is what is usually done, as it is easy and convenient. If identity verification later on from a polled ballot is the sole criteria of a fair election, then most countries that do not record a fingerprint, or copy DNA onto a ballot (which is almost all countries), do not have a fair election process. Most countries do not record aspects of voter identity onto the ballot. Voter identity is kept pooled at polling stations. Indian EVMs just record counts and polling stations keep voter record but only dark ink on the thumb as a detriment to multiple balloting. What they do is make sure counts are accurate (thanks to EVMs) and the delay in between two votes on an EVM makes it a very tiring exercise to stuff ballots at a PS anyways. Polling station registrations being low (800 to 1,200 eg) keep ballot stuffing as less than successful electoral fraud mechanism. Angels helping military dictators had trouble in it too. It would be a fair assumption to say that no more than 10,000 ballots can be stuffed and the rest has to be done through count manipulation. See a recent case for example. Appointing fake presiding officers and count manipulation at polling station proved a fraud exercise. The culprit ended up being caught easily as the high correlation b/w the NA and PA candidate counts did not exist and testimonies proved using POs to manipulate results.

You cannot expect an election where the identity of every voter is established post-voting from the ballot paper. What you should expect is a strong mechanism which makes sure that ballot stuffing is rare (candidates’ polling agents being the bulwark against that), where voter suppression and intimidation is not omnipresent and where total counts do remain the same, from when the left the PO at a PS to when the RO compiled them. The last being an easy and most common rigging mechansim, electronic voting of the cheap, reliable kind that India employs is the answer. You do not need error prone, unreliable, expensive ‘biometric’ voting where each identity is recorded with a ballot. You need efficiently trained polling staff, reliable voting machines, non-tamperable counts and candidates who contribute polling agents who record violation of the process. It makes no sense that an election be nullified if one out of a million votes is questionable, or a dozen are found to have been manipulated. In a utopia, it would matter but in real life, it’s statistically significant numbers that matter. I can expect to have this section quoted out of context as proof of hypocrisy or such thing but if you got 45 votes and the winner 120,000, there is little public interest in nullifying the election because you called the authenticity of 100 votes into question.

As I mentioned earlier, I’d come back to the case of Tariq Malik in the end. His removal has been seen as ultimate proof of subversion of the electoral verification process by most. The perception is that he was going to expose rigging in NA-122 and he was removed to prevent that. But it was not upto him to do a verification. He could only do so if asked by an election tribunal, and if the case is pending a stay order in the high court, or languishing in a tribunal, the Chairman NADRA can’t just obtain the election material from the ECP and produce a report on his whim. Even after his removal, the sham exercise of fingerprint verification has kept ongoing. It has not stopped, has it? But the question remains, why was he removed? His elevation to Chairman NADRA by the PPP, in violation many rules forgotten, from what I can gather, and some of it is publicly clear, a positive press hungry Tariq Malik pressed on with the ‘fingerprint verification’ exercise to the ECP immediately after the elections. The Rs 90mn for the ‘software upgrade’ acquired (and many, many shady deals before), you produce reports that lead to positive press (XPOZED RIGGING). He was going press hungry with it, and a panicked PML-N (proof of unnecessary panicking is no longer needed), had him fired. He was whisked out of the country with the promise that he won’t speak to the media, which he violated lately, and is thus being hounded by the FIA for the scams for which he was not processed earlier. Using systemic errors, random errors, system biases and technical limitations to blackmail the government and get good press resulted in overnight removal (which actually was considered for quite a while and not overnight like it’s been portrayed). One of the non-financial clean-washed sins includes having acquired a false identity. Besides having a second citizenship (Canada), he had an official Pakistani passport, a MRP for a second identity and an old style passport for under the radar travel. By virtue of his post as Chairman NADRA, he was the Registrar General of Pakistan, responsible for all identities, and was himself involved in identity fraud! Nonetheless, it does not mean that he could not have been fired to put a stop to some real fraud being exposed. It could and might have been true and facts need to come to light.

I’d like to end this by saying that candidates’ Polling Agents are the first and most important barrier to fraud. They oversee the election at polling station level, seeing any ballot stuffing, voter suppression or any other fraud going on. They receive Form XIV of their polling station (it is posted publicly outside the PS after the count too) and Form XIVs are received by the election agents from the RO. They should already have (or can acquire them without too much difficulty) what’s necessary to expose rigging in the form of count manipulation too . If they did not acquire these documents, or have not tried to do so, they have failed their own candidacy.

FAFEN has suggested the simple, reliable and effective way of electoral audit. Form XIV (count at polling station) and Form XVI (complied count by RO) be used to expose the manipulation that is most commonly done. This will expose rigging in almost all of the cases where it existed. If a statistically significant number of votes were manipulated, the election result should be nullified, culprits should be punished and re-polling be done. Ballot stuffing itself is identifiable using nifty statistical techniques. Open data of PSs will enable all parties to do that. See Afghanistan’s vote audit right now to see how simple and laughably obvious count manipulation and voter fraud at a large level is. What you need is cross-checking of Form XIVs and Form XVIs as primary mechanisms of an electoral audit, to be followed by recounting in suspicious cases and use of statistical techniques on a larger level to show evidence of widespread fraud. Institutionalized electoral rigging is blatantly obvious.

About Shahid Saeed

Shahid likes to waste time reading history.

Comments are closed.