Daily Times

The provisional results for Pakistan’s sixth population census, carried out in 2017, are out! Some significant takeouts from the Census are that Pakistan’s men are more than women; that there are 105 men to 100 women. Moreover, Sindh is the most urbanised province having 50.02 percent urban population. This census is very important for Pakistan: its results will impact the provinces’ financial share in the National Finance Commission award and in provincial job quotas. It will be responsible for redrawing battle lines or, in simple words, redistricting electoral constituencies for upcoming general elections of Pakistan. Census 2017 results are already generating heated debates. They are likely to remain doing so for a long time to come.

Census data is government-collected data. Notwithstanding right of logical debates, there is a difference between veracity of numbers suggesting 1 in 5 women advocating use of a particular fairness cream for miraculous results, and government census numbers for population increase over past 19 years. We have every right to roll our eyes over fairness cream statistics, but the attention and emotion is quite different when we question that Pakistan’s population ballooned 57 percent from last 1998 census. The difference is that the later claim

does not come from a private company but a formal government. Statistics gets its name from the discussions of state, called ‘statisticum collegium’ in Latin. The aim has been to better measure the population to better serve it. That is why we need governments conducting surveys and generating numbers, drawing trends and making policies based on them. But we need to move beyond blindly accepting or rejecting any statistics, or even their offered analyses and official interpretations. We all need to learn the skill to spot bad numbers.

Numbers can make us wary and skeptical, but one should always be able to differentiate between reliable and unreliable numbers. This is all about swimming in huge pool of big data. Large statistics look formidable and impressive, and most of the world’s population is in equal awe and fear of numbers. Scientists and policy makers throw averages and other numerical quantifiers at us, because we cannot deconstruct them so we blindly accept them. This happens due to quantification bias: which is the natural human preference of measurable over immeasurable. But the fact is that numbers can lie, and more often, they do. The climate change debate is a clear example of manipulation of numbers by people denying objectively measurable physical data. Now how can a simpleton understand if the census, or any big data results for that matter, are true or not?

Numbers can make us wary and skeptical, but one should always be able to differentiate between reliable and unreliable numbers. Large statistics look formidable and impressive, but the fact is that numbers can lie, and more often, they do

Remember! Data is not important, facts are important. They convey realities. For example, data tells us that number of men is more than women; a fact is that around 1.2 million female fetuses were aborted between 2000 and 2014, and that the man to women ratio of 1.05 for that period is a genuine statistic. Is this numerical proof that Pakistan’s strong preference for sons is a fact?

In order to understand the census, or any big data findings, we need to ask a few fundamental questions. The list can be longer but some basics need to be always there. The first question should be “How was the data collected”?

In case of Pakistan’s census we know that Pakistan Bureau of Statistics (PBS) was responsible for the exercise. PBS has collected data manually while using Optical Character Recognition (OCR) technology for conversion in machine readable date. OCR is not extremely accurate and the data was thus also double-checked manually by the data operators before conversion to machine readable format. Manual collection means that at many occasions the data was orally collected and filled into forms by the teams. This methodology can create bias in data. How was this bias accounted for? And how much of the data been affected by it needs to be questioned?

The second question that needs to be asked is “how is the data being communicated?”

They say devil is in the details, and it usually is. We have been presented the larger trends like Sindh is more urbanised, or Islamabad’s rate of urbanisation has stalled. This is selective communication of the census data by the media. It is not very wise to draw conclusions from this alone. The axes are everything; once you change the scale, you can change the story.

Lastly, in the list of non-exhaustive questions to spot bad numbers is “Where do I see myself in this data”. This is a very interesting aspect of data. This shows that the data is representative and is not skewed or missed certain segment of the population, that is, you.It may not be applicable readily to your immediate self, but you should be able to relate to the data. If you are an above 30, married female, try finding out how many of the females like you filled in the specific form. You may be able to see a trend that you can relate to.

Whether it is census data or any other big data, being a layman shouldn’t stop anyone from differentiating between good and bad data. Training one’s self by verifying the evidence behind big claims is the key. Evidence-based thinking is not only required in assessing big data, but for evaluating every aspect of life.

The writer is a policy practitioner, an Oxford public policy alumnus and Oxford Global leadership initiative fellow

Published in Daily Times, August 28^th 2017.