All eyes on the 2021 census

The Central Bureau of Statistics has belatedly begun the enumeration for the census, after postponing it several times due to the pandemic. The enumeration phase began after the completion of a prolonged pre-enumeration process, which consisted of tasks related to planning and developing new questionnaires. 

Many political and social contexts have changed since the last census in 2011. The country has been re-structured into federated units with 7 provinces and 753 local units. There is a genuine demand for data from these units. The CBS has addressed this in a few ways, some substantive others procedural. These and other changes from the last census or their lack of are bound to affect scope and quality of data generated. 

In this essay, I assess the quality of Nepal’s latest census, by discussing how the census is adapted to the needs of the federal context and then move on to analyzing the questionnaire. Finally, I will discuss the implications of sensitive questions that are used to enumerate ethnicities, languages, sex-gender-sexuality, religion, and people with disabilities. While these are only a few aspects of the whole census process, they provide a good peek into current census’ advances and limitations.

Adapting to the federal system

The census will administer a total of 54 questions throughout the country, which is supposed to provide disaggregated data for all governing units — all the way to ward level. In 2011, a long form-short form approach was used where only 1 out of 8 households was given a sample of all the questions, while the rest were administered an abridged version. The sample was then extrapolated to make population estimates. Since the current census will administer all questions to all the households, it takes longer to enumerate and will add to the workload of enumerators and supervisors. If this workload is not managed by hiring enough human resources, the data quality will suffer. In comparison to the 2011 census, the CBS has increased the number of enumerators from 30,000 to 38,000 and increased the days to enumerate from 12 to 15. These adjustments cater to the increased workload and to the fact that the population has increased in the past 10 years. However, a longer questionnaire will add administrative load throughout the process. As such, there are concerns about how the increased workload will impact data quality. The value of providing disaggregated data till the ward level is enormous but it remains to be seen if CBS can pull it off without compromising data quality. 

Another way the census is trying to adapt to the federal system is by employing as many local supervisors and enumerators as possible. The logic follows that a local enumerator will be familiar with local contexts thus ensuring data accuracy. For example, a local enumerator is less likely to miss out on a household from his/her own community. They are familiar with the local languages and ethnicities and can correctly input the data. However, this method has its limitations. For example, local enumerators and supervisors are embedded within local power matrixes and thus could influence data collection efforts. More importantly, the quality of data collection also depends upon the quality of training for supervisors and enumerators and the quality control that CBS can ensure through the supervisors and local/regional offices. It is in this aspect that Nepal’s earlier census has been particularly weak, with training not meeting the necessary standards and supervisors only making procedural checks rather than concrete ones. In this census as well, supervisors are suggested to do spot-checks and back-checks at the beginning of the survey, but they are not provided with concrete targets for the number of such checks to be carried out. Without specific targets, the quality control cannot be ensured across the country. Reports of alleged negligence by enumerators have been widespread.   

CBS has also made provincial and local committees and provided them with coordinative functions which is expected to be conducive for data collection but could also allow for a source of political influence and contestation in the process. Similarly, the census is administering a “community questionnaire” to the ward chairpersons. This questionnaire from a data collection point of view will be ‘guesstimates’ by the ward-chairperson. The CBS’s rationale seems to be that this would help in ascertaining data accuracy but data collection through key-informant-interview will be fraught with respondent bias. Instead, it will add to the workload of the supervisors. At face value, this appears to be futile as the data accuracy from a household survey should not be cross-checked by a key-informant survey. It would have been better to invest in ensuring quality checks in the household survey rather than relying on key-informant surveys for ascertaining data accuracy. 

Evolution of the questionnaire 

While preparing the questionnaire, the CBS did consult a broad range of stakeholders, including at the sub-national level. This reflects in the final questionnaire. It captures a wide range of indicators. However, this makes the questionnaire lengthy. Many of the indicators, which could have been captured through regular sample surveys, have been added to the census while the concern for respondent attrition (on such a large-scale survey) has been kept secondary. 

Data generated out of ambiguous questions can make census estimates unreliable. Most questions in the final questionnaire are straightforward, but, in my opinion, several key questions, including sensitive ones, lack definitional clarity expected in a census. There are deeper problems that the CBS needs to address but part of it is due to the way the pilot test was run. One questionnaire was piloted in more than 10,000 households and the final questionnaire only underwent a few minor phrasing adjustments. The fact that the CBS tested only one pilot questionnaire shows us where the problem lies. 

Perhaps the best illustration of how the pilot census was not properly utilized is exemplified in the way questions related to economic activities were changed in the final questionnaire. In the pilot, the section intends to answer if an individual was unemployed, engaged in agricultural activities or other economic activities in the past seven days; followed by the reasons for unemployment in the last seven days, labor participation in the past 30 days and willingness to seek a job or a business provided the opportunity in the next 15 days. The question follows ILO standards and can yield comparable statistics on unemployment, labor force participation, engagement in the agriculture sector, among others. The recall period of 7 days, 30 days and projection of 15 days do not place much respondent burden. However, these questions have been completely upended in the final questionnaire, evidently without any testing or at least the record of having tested. The final questionnaire also puts a huge respondent burden with a 12-month recall period. Number 31, 32, 36 and 37 of the questionnaire do not follow comparable survey standards. Why this change was made will perhaps come to light once the census releases its report on the pilot census, but one thing is clear: CBS needs to conduct pilot surveys more systematically — with a clear objective to vigorously test the questionnaire. It can carry out questionnaire tests on a small scale in several purposively selected places, and repeated rounds of questionnaire tests must be carried out until a satisfactory questionnaire evolves. CBS can even experiment with translated questionnaires during such pilots, which can provide the basis to administer future surveys in more than one language.

Politics and the methods of sensitive questions 

A census is as much a political exercise as it is a statistical one. Count estimates of various ethnicities, languages, religion, people with disability, and gender and sexual minorities have important meanings especially since our political system purports to be proportionately representative. By extension, these estimates also shape the political discourse on inclusion and marginalization, at least for the next decade. This requires CBS to maintain a delicate balance in its methods so that there is a general agreement on the census estimates of sensitive issues. The methods used in sensitive questions have seen some changes from the last one but are yet to strike that delicate balance. 

Take for example the question counting caste-ethnicity (jaat-jaati). The enumerator’s manual states that jaat refers to the surname in hindu caste hierarchy, while jaati refers to the indigenous ethnicities. However, since the same question attempts to answer both jaat and jaati, it can get very confusing to respond to it. Especially in Nepal, where many ethnicities and castes share similar surnames and surnames are commonly understood as denominators of ethnicities and caste, often under-trained enumerators have presumed the caste and ethnicity of their respondents. The recorded responses then will depend upon the varied interpretations of different enumerators. This makes the data on caste and ethnicity very unreliable. 

Similar problems persist while enumerating languages. Language has been a major tool for homogenizing the Nepali identity ever since the panchayat system made Nepali a national language. This unfortunate piece of history has made counting languages quite difficult, with several communities vying separate enumeration for their own languages and dialects, even when many of the languages might closely relate to one another. For an enumerator, it can get quite difficult to distinguish between the two unless one is adept in that tongue. CBS has attempted several renditions to capture this in the past and has again innovated in the current census. The census asks for a person’s mother tongue, second language and ancestors’ language; however, the definition of second language and ancestors’ language can easily be misunderstood. Indigenous communities fear that it is a tactic on the part of the state to garner support for the Nepali and Sanskrit language by recording it as the majority individuals’ second and ancestors’ language respectively. It is unclear if the question will yield this result, but such fear is founded on the historical injustices carried out by the state against different linguistic groups. The fear is further aggravated because the definition of these questions is prone to misinterpretation.

On the other hand, CBS can get drawn into political fights even when its methods are modestly impartial. This is true when it comes to the question of enumerating religion. In comparison to jaat-jaati and languages, enumerating religion is simpler. The Constitution also provides a person with the right to determine one’s religion, which means the census can simply let everyone determine their religious identity. However, several religious groups in a bid to maintain their higher composition have challenged innocuous categorization made by the CBS. This risks igniting religious extremism once the census results do not meet such groups expectations. 

Similarly, the question and responses to enumerate people with disabilities (PwD) is also not likely to yield a different result than the conservative estimate that came out in 2011. It had estimated that less than 2 percent of the population were PwD in Nepal. Many developed countries with higher quality disability statistics tend to have 10-15 percent people with disabilities. The current question to enumerate PwDs is inclusive of most forms of disability, but several overlapping responses are included. For example, one person can be included in several responses — ‘multi-disabled’, ‘deaf’, ‘related to voice and speech’. The question in its textual form is very leading; it asks, what type of disability does ‘name’ have? It risks making the respondent hostile to the enumerator. In fact, even the enumerator manual seems to identify this problem, but necessary amends were not made to address it. Unlike for enumerating languages and ethnicities which require extensive deliberation, there is already an international standard for enumerating people with disabilities. The census could have used the method developed by Washington Group on Disability Statistics which includes six quick disabled-sensitive questions. 

For enumerating gender and sexual minorities, the CBS only includes an option of ‘third gender/other’ in the household enlisting form. The explanation in the supervisor manual for this categorization appears to mix both gender and sexual minorities in the same bracket. CBS needs to correct this oversight in the future survey exercise. However, enumerating gender and sexual minorities has been a challenging task for many countries. Several concerns come up across the world while enumerating gender and sexual minorities including their right to privacy, and enumerator and respondent’s prejudice. In countries where more thorough attempts were made, the final enumeration was very low. Even in Nepal, the 2011 census only enumerated 1,500 gender and sexual minorities. Rather than as a data collection exercise, for now, the attempt to enumerate gender and sexual minorities can be understood as a way to ensure their political representation in a key governance process. The current census has ensured representation in one way but has failed to maintain a nuanced definition of gender, and sexual minorities. 

Conclusion

Census 2021’s attempts to provide detailed, accurate and disaggregated data all the way to the local level must be commended for its ambition. However, concerns remain over its capacity to pull off such a large survey without compromising the quality. This concern naturally stems because the conception of questionnaire remains problematic, with lack of definitional clarity vexing sensitive questions for enumerating jaat-jaati, languages and people with disability. Some of these problems can be resolved by simply following the templates that are available internationally, while for enumerating languages and jaat-jaati a more thorough research and consultation with communities is necessary which ought to be done in the intermediate period between two censuses, not just when the census process starts. Doubts also remain over the rigor with which the final questions were tested, evident from the haphazard changes made to the pilot questions on economic activities. The results of the 2021 census need to be interpreted considering these limitations. It will also be important to learn from these shortcomings for delivering higher-quality surveys in the future. 

Disclaimer: This article was written based on the Census 2021 research and publication by Democracy Resource Center Nepal which this writer co-authored. Subedi currently works at The Asia Foundation, supporting the subnational governance and dialogue program.  The analysis in this article is the writer’s own and might not represent either organizations in any shape or form.