Abstract: Dr. Abusaleh Shariff has written a few articles and responded to questions in an interview stating that Muslim voters are being excluded from the electoral rolls and that the Muslims are losing interest in Electoral Politics. His views are based findings from his research. My analysis following similar methods as Dr. Shariff shows that Muslim representation in the electoral rolls is proportional to their population and that they participate in elections with the same vigour as the non-Muslims.
“Electoral Exclusion of Muslims Continues to Plague Indian Democracy,” is the title of a paper published in the Economic and Political Weekly (EPW), Vol. 53, Issue No. 20, 19 May 2018. See. The abstract states,”…nearly one quarter of Muslim adults in Karnataka were out of the electoral rolls [as against] about 15% of all adults … Evidence of non-listing of Muslim electorate in large proportions is found in other states as well …”
On 30 October, my attention was drawn to the paper in EPW. While I was reflecting on it, I found a few more stories based on this article. Khalid Saifullah, Abusaleh Shariff, and Mohsin Alam Bhat wrote a similar story in Times of India, Edit Page, 04 November 2018, adding a few more arguments. Dr. Shariff’s interview with Scroll.in was more damning. See. An accusative version of the article also appeared in Vartha Bharathi epaper. Economic Times editorial wrote referring to the Times of India article.
Center For Research and Debates in Development Policy, New Delhi (CRDDP), has a website named missingmuslimvoters.com. They do research on missing Muslim voters and CRDDP urges the community leaders, politicians and philanthropists to support this cause to politically empower the Muslims of Karnataka and seeks donations.
Mr. Wazaht Habibullah, the former Chief Information Commissioner of India, took up a case with the Chief Election Commissioner (CEC) based on Dr. Shariff’s paper. The CEC referred the case to Chief Electoral Officer (CEO) of Karnataka. The current status of the case is not known. Some learned and highly respected people appear to be supporting Dr. Shariff’s views.
Source of Data
CEOs of states and UTs publish the electoral rolls of their states/UTs at their websites in PDF format, generally twice a year. In election years, they publish the rolls 3 or 4 times. I have studied the electoral rolls of 13 states/UTs and have been analysing every version of the electoral rolls of Delhi and Bangalore since 2010. This gives me a fairly good understanding of the Electoral Roll Management System (ERMS) of the Country. However, I had not checked for any community bias in the system.
The above findings by Dr Abusaleh Shariff et.al. are disturbing. If the findings are true, it is sad and I would work with people trying to correct this situation. It is equally bad if the findings are not consistent with the results of another analysis of the same data because misleading observations of such sensitive matters can affect communal harmony.
The authors state that the findings are based on electoral roll data published by CEO, Karnataka. Because I have 19 versions of Electoral Rolls of Bangalore District published between April 2012 and October 2018, it is easier for me to repeat their analysis to test the veracity of Dr, Shariff’s findings.
With Dr. Shariff and his IT specialist, I have shared my sample data, references, and source code used for the analysis. I have offered to work with them on this analysis task.
As stated in his interview with scroll.in, Dr. Shariff had chosen 25 constituencies of Karnataka for his research. He has stated, “Any statistician will tell you that an 11% sample is good enough to understand an all-pervasive event in the universe of 224 constituencies.”
As reported, Karnataka had 49,682,351 voters for the assembly elections in May 2018. My sample consists of about 90,00,000 voters of 28 constituencies of Bangalore District, with 18.12% of Karnataka voters.
Table-1: Comparison of Samples
|Dr. Shariff’s Sample||PG Bhat’s Sample|
|From rural Karnataka||From urban Karnataka – only the voters of Bangalore|
|The sample has 11% of Karnataka voters||A larger sample with 18.12% of Karnataka voters|
|1. The population of Bangalore may not be much different from the rest of Karnataka in the % of Muslim voters and the voters’ beliefs and behaviour.
2. Data quality of electoral rolls of Bangalore is poorer than that of the rural Karnataka as agreed by the CEO in some meetings.
3. I consider that the two samples represent the factors under analysis equally.
Table-2 below has a summary of the findings in the two analyses – one by Dr. Shariff and the other by me. Details of each item are explained in succeeding paragraphs.
Table-2: Differences in the factors observed by Dr. Shariff and PG Bhat
|Dr. Shariff||PG Bhat|
|1||18% of applications for registration were rejected with a reason that the applicants are not citizens of India.||As understood from the CEO’s office, the report generator software had an error which reported wrong reasons. The software has since been corrected and verified.|
|2||20% of the Muslims are excluded from the electoral roll.||The electoral rolls have 12.94% Islamic names, close to their population %. Voters are not excluded from the electoral rolls with any communal bias.|
|3||Muslims are slowly withdrawing from electoral politics.||Muslims register as new voters with the same zeal as non-Muslims.
Voter turnout% is higher in booths where the % of Muslim voters is high.
Muslims seem to participate in electoral politics more vigorously than non-Muslims.
|4||Single-voter household indicates the exclusion of voters. 20% of the Muslim households have single-voters whereas only 12.3% of the non-Muslim households have single-voters.
This implies exclusion of Muslim voters in Karnataka.
|1. % of single-voter Muslim households is about equal to the % of Muslim voters. Similar is the case with non-Muslims.
2. Electoral rolls do not have dependable house addresses. Many people living in one house are shown in different houses.
3. We cannot group the electors by houses with unreliable data on houses.
We cannot conclude that Muslims are excluded from the electoral rolls based on the counts of single-voter households.
|5||Up to 15% of the total electorate of around 130 million adult citizens is missing from the electoral rolls.||Census report states that 41.1% of Indian population is below 18 years. CEO-KA reports an elector/population ratio of 72.8%. This indicates a bloat of 15% – not a deficit.
It is not possible to estimate the count of unregistered voters based on electoral roll data rife with illegal entries.
Applications Rejected on the Ground of Applicant Not Being Indian Citizen
Dr. Shariff: “… during 2017-18, a high 62% were rejected — as many as 18% were denied on the ground of not being Indian citizens and 24% for unspecified reasons.”
Analysis & Discussion:
It was later found out that applications which were rejected for reason ‘Shifted Residence’ were shown as ‘Not Indian Citizen’ in the report generated in response to queries on CEO-KA website. This was due to an error in the report generator software, which has since been corrected. Subsequent checks show these records as rejected for having shifted residence and not for non-citizenship.
Between January and October 2018, only 10.61% of the applications for registration as voter were rejected in Bangalore as seen from CEO-KA website. This is far lower than 62% rejections observed by Dr. Shariff.
Thanks to Dr. Shariff’s observation, the error in software was corrected promptly.
Islamic Names in the Electoral Rolls
Dr. Shariff: “I realised that the number of Muslim voters has declined over the years. This realisation became my motivation to study Karnataka’s electoral rolls. … A large number of Karnataka’s adults are not on its voters list. This number is particularly high for Muslims – 20% or 13 lakh Muslim adults are not on the electoral rolls. In comparison, 12.3% or 53.2 lakh non-Muslims will not be able to vote for the same reason. This finding should be seen in the context that Muslims constitute just 13% of Karnataka’s population.”
Analysis & Discussion:
Classification of voter records. From The Dictionary of Islamic Names and about ten websites giving Muslim baby names I collected the words in Islamic names. From each voter record, I checked every word in voter name and relative name to see if it is in the collection of Islamic name words. The % of words in the voter name and relative name identified as Islamic name word is the score for the record being of a Muslim voter – ranging between 100% to 0%. Where the score is above 50%, the name is considered Islamic. Manual verification of this simple classification in 100 randomly selected electoral rolls showed that the classification is acceptable. False positive cases were lower than false negative cases, i.e., wrong classification as Islamic names was lower than wrong classification as non-Islamic names. A more detailed classification scheme can be employed. However, for the current analysis, a small margin of error is considered acceptable. This classification is similar to the one used by Dr. Shariff’s research team.
The electoral rolls of Bangalore used in the latest Assembly Elections in May 2018 have 12.94% Islamic names. It may even be a little higher considering that we have a little higher false negative cases than false positive cases. This is very close to the approximated 13% Muslim population in the state. Dr. Shariff’s claim that 20% of the Muslims are excluded from the electoral rolls compared with 12.3% of non-Muslims is not true in Bangalore. We cannot conclude that a larger number of Muslims are excluded from the electoral rolls of the state.
Do Muslims Participate in Electoral Process?
Dr. Shariff: “… It was a feeling that Muslims are slowly withdrawing from electoral politics and that their voting behaviour has changed from the past.”
Analysis & Discussion:
Registration as Voters. The CEO-KA publishes lists of applications for additions, deletions, and modifications and their status on his website. Between February and May 2018, 13.38% of the additions are of Islamic names, a little higher than the % of their population. The Muslims are as engaged as others in the electoral politics especially on the eve of an election.
Voter Turnout. Forms 20 give the election results at booth level, from which we know the total votes cast per booth. From the electoral rolls, we know the number of voters in each booth. From these two data, we derive voter turnout% for each booth. Having classified the voter names as Islamic and non-Islamic we also know the % of Muslim voters in each booth. If the Muslims are disengaged from the electoral process, turnout % should be lower where the % of Muslim voters is higher and vice-versa. This would be a negative correlation between the % of Muslim voters and voter turnout%. Studies show a clear positive correlation when the % of Muslim voters is above 25%, indicating that Muslims certainly participate in electoral politics, even with more enthusiasm.
Muslims do not seem to be disillusioned with the electoral process.
Appendix-1 has scatter plots to visualise the correlation between the % of Muslim voter in parts and voter turnout%.
Single Person Households
Dr. Shariff argues strongly that single-voter households indicate that other eligible electors from the house are excluded from the electoral rolls and with this logic states, “… the incidence of exclusion among Muslims is nearly 24 percentage points more than among All-Others…” He has also mentioned that CRDDP estimated net single person voter households as a percentage of total voters (%) is 20% for Muslims and 12.3% for others.
Analysis & Discussion:
From the electoral rolls of 28 Bangalore constituencies, I calculated the following for each part:
- Voters in the part.
- Count of voter records with Islamic names – Muslim voters.
- % of Muslim voters in the part.
- Number of houses in the part – unique value of (section number + house number)
- Number of houses with single
- Number of houses with single voter where the voter is Muslim.
- % of single voter Muslim households.
- Variance = item7 – item3. Positive variance indicates possible higher exclusion of Muslims and negative variance indicates possible lower exclusion of Muslims.
Finding: In 8,286 parts, the average variance is 0.20%. Dr Shariff’s statement that % of single person households is 20% for Muslims while it is 12.3% for others does not seem to be correct in Bangalore constituencies representing more than 18% of Karnataka voters. Dr. Shariff’s conclusion is for the entire Karnataka.
In Bangalore constituencies, the % of single-voter Muslim households are proportional to the % of Muslim voters in the part, similar to non-Muslim voters. This differs from Dr, Shariff’s observation.
One reason for a large number of single-voter households is wrong addresses in the voter records. There are many cases of members living in one house being shown in different houses. E.g., the corporator of BBMP ward 162 actually has 3 voters in her house. They are wrongly shown in 3 different houses. When 3 members of a household are shown in 3 different houses, instead of one household with 3 members, we may get 3 single-voter households. There are many examples where voters in one house are shown in different houses.
Dr. Shariff: “Our data for Karnataka shows that there are 66.2 lakh households in Karnataka, of which 13 lakhs are Muslim and the remaining 53.2 lakh All-Others.”
Analysis & Discussion: Dr. Shariff’s finds that 19.64% of the houses in the electoral rolls are of 13% Muslim population. However, in Bangalore, the % of Muslim houses compare well with the % of Muslim voters. House addresses in the electoral rolls are highly unreliable. Analysis findings based on them would not be reliable.
Low Elector/Population Ratio
Dr. Shariff: “Up to 15% of the total electorate of around 130 million adult citizens is missing from the electoral rolls.”
Analysis & Discussion:
Census India report states that 41.1% of Indians are below 18 years old. Hence, Elector/Population (E/P) ratio higher than 58.9% would mean fake and illegal entries in the electoral rolls – even if every eligible citizen is registered as a voter. In a press statement on 10 Oct 2018, CEO-KA stated that the E/P ratio for Karnataka is 72.8%. The electoral rolls are rife with duplicate entries. Many dead persons or the ones who have shifted residence are not removed from the electoral rolls.
With about 15% bloat in the electoral rolls, it is not clear how Dr. Shariff could conclude that the rolls are missing 15% of the voters based on the data from the electoral rolls.
In the past, there have been several cases of deletion of voters without due diligence. E.g.,
- In 2012, CEO-KA deleted 13.5 lakh voters out of about 65 lakhs voters of Bangalore. In response a PIL, the High Court of Karnataka directed the CEO to restore such voter suo-motu.
- In April 2014, admitting that there was indifference on the part of poll officers in Maharashtra, Election Commissioner HS Brahma told Economic Times that the large number of deletions from the electoral rolls in the state was unfortunate and the poll watchdog regrets the error.
- In December 2018, Shashidhar Reddy complained that about 68 lakh voters were missing from the electoral rolls of Telangana. CEO, Telangana, apologised for the lapses.
In the light of such occurrences in many states, it is important to investigate the reasons for such repeated lapses and also to see if there is a community bias. We have to study deeper to find the full facts.
Conclusions – Large Variance in Findings Between Two Analyses:
By repeating the analysis using the same approach as Dr. Shariff, I find contradicting results. I have offered to work with Dr. Shariff on this analysis and have shared the data, references, algorithm and source code. Exploring further – refining the analysis methods and working with different sets of data samples – may reveal truths we do not know. Inclusive democratic processes are very important and we cannot ignore symptoms to the contrary.
The inescapable conclusion is that Dr. Shariff’s analysis could be perceived as coloured based on extraction of convenient data
Appendix-1: Scatter Plots
Scatter plots are used to plot data points on a horizontal and a vertical axis in an attempt to show how much one variable (dependent variable) is affected by another (independent variable). Each row in the data table is represented by a marker whose position depends on its values in the columns set on the X and Y axes.
The relationship between two variables is called their correlation. If the markers are close to making a straight line in the scatter plot, the two variables have a high correlation. If the markers are equally distributed in the scatter plot, the correlation is low, or zero. However, even though a correlation may seem to be present, this might not always be the case. Both variables could be related to some third variable, thus explaining their variation, or, pure coincidence might cause an apparent correlation.
The following scatter plots with trend lines (line of best fit) have Muslim Elector % in X-axis as the independent variable and the turnout % on the Y-axis as the dependent variable. Data points, represented by dots on the plot, are the voter turnout% in a booth with the % of Muslim voters shown on the corresponding position on X-axis. The positive gradient of trend lines shows that voter turnout% improves when there are more Muslim voters in the part.
Fig-1: Voter turnout% against % of Muslim voters in the part
Fig-2: Voter turnout% against % of Muslim voters where Muslim voter% is between 25 and 50
Fig-3: Voter turnout% against % of Muslim voters where Muslim voter% is more than 50
- If Muslims are not engaging in the process, we should expect voter turnout percentage to be very low in the areas with the higher Muslim population. It appears this is not so.
- Further, there are almost no Muslim dominated areas with very low turnout. It doesn’t seem like those areas are not engaged in the voting process. In fact, the problematic areas (low voter turnout) are areas with low % of Muslims.
- The big question is why is the electorate not turning up in some areas relative to the others. The variance in voter turnout is highest at low values of Muslim electors. So there is some other factor (and NOT Muslim elector %) that may better explain why some areas have low voter turnout than others.
- We can run a regression of voter turnout % against all known attributes of the electorates. These include economic indicators (income), social indicators (caste %), education, and other factors. We can see what are the economically and statistically significant variables. It’s unlikely to be Muslim voter % based on the scatter plot.
Appendix-2: Other Points
Dr. Shariff: ” I have 40 years of experience in data collection and data mining. All the protocols through which any data has to pass were followed, including cleaning and setting. The procedures we followed were labour-intensive and often, meticulous matching of individual cases became necessary.”
Analysis & Discussion: Despite the experience and rigour, we find that some outcomes of the research do not match with the reality.
Dr. Shariff: ” Obviously, I wish to know the social identity of the missing voters in the All-Others category. My sense is that it would be marginalised groups like Dalits. I say this on the basis of the Muslim community being a marginalised group ”
Analysis & Discussion: Such statements, without evidence, can create social disharmony.
Dr. Shariff: “We did bring the issue of exclusion of Muslims from the voters list to the notice of the state’s minority commission and the Karnataka government secretary who liaises with the Election Commission. We also wrote to the Election Commission of India and the chief election officer in Karnataka”
Analysis & Discussion: Dr. Shariff’s analysis repeated with another sample of a larger size gives different findings. I hope that Dr. Shariff’s complaints won’t be taken on their face value without a deeper and more complete analysis.
” It could be a demand and supply kind of problem. For instance, over the years, Muslims have come to feel that it does not matter whether they vote. For instance, the Bharatiya Janata Party has not fielded a single Muslim candidate in many recent elections. It also says it does not want Muslim votes. This kind of discourse discourages or holds back Muslims from enrolling themselves in the voters list. This could very well be the psychology of other marginalised groups too.
Second, those entrusted with preparing the voters list simply go to a chauraha [crossroads], point at homes and ask who their owners are. Normally, the name of the person heading the household is noted and the other members left out. They do not conduct a house-by-house survey. Guiding the “enrolment officer” is often a person from a dominant social group.
Third, there could be a systemic bias against Muslims, a bias so strong that leads to the exclusion of an entire locality or area in a constituency.
The demand side of the problem is about the system not motivating or encouraging them to vote. There are politicians who suppress certain social groups from coming out to exercise their franchise. But the same politicians also motivate voters from their own communities to enrol and vote. Unfortunately, the Election Commission does not seem to have proactively stepped in to prevent such a manipulation of the electoral system.”
Analysis & Discussion:
The first point is not true with the electoral rolls of Bangalore – a sample larger than the one chosen by Dr, Shariff. % of Muslim voters seem to be close to the % of Muslim population. Also, Dr. Shariff’s statement gives a political colour to the argument.
The second point is different from my experience with the electoral registration process with which I am closely involved for the past nine years.
The data does not show any bias against any community though such a possibility is mentioned in the third point.
The last point is different from the fact because we see that Muslims turnout to vote better than other communities