What Does the Indian Parliament Discuss?
An Exploratory Analysis of the Question Hour in the Lok Sabha

Among a total of 298,292 questions, the number of questions asked in each term of the Lok Sabha is as follows: 13th: 73,531; 14th: 66,371; 15th: 79,401; 16th: 78,989. We found out that over the span covered in this dataset, Lok Sabha always met for more than 50 sittings in a year, except in 2004 (48 sittings) and 2008 (46 sittings). This correlates with the lowest number of questions asked in 2004 (9398 questions) and 2008 (9851 questions). Two of the three sessions in 2004 and all sessions in 2008 are included in the 14th Lok Sabha. Although the fewer sittings are highlighted in many news reports [Anuja, 25 Nov 2020], we did not find mention of its impact on Question Hour, which we clearly identified above.

Unlike the previous Lok Sabha terms, in the 16th the ruling alliance asked more questions than the opposition. In that term, they had a historic $65\%$ share of the House. Since the numeric strength of the ruling alliance is, by rule, higher than that of the opposition, we normalize them to get an idea of participation from each side had the number of representatives from either side been equal. We believe this will afford a fairer comparison between their participation. Let $R_{n}$ and $O_{n}$ be the number of members of the ruling alliance and the opposition, respectively, and $R_{q}$ and $O_{q}$ be the number of questions asked by the ruling alliance and the opposition, respectively. Then $R_{pp}=\left(R_{q}\times\frac{O_{n}}{R_{n}+O_{n}}\right)$, $O_{pp}=\left(O_{q}\times\frac{R_{n}}{R_{n}+O_{n}}\right)$. As seen in Figure 2, in the transformed space, the opposition still asks more questions than the ruling alliance, matching the expectations of a healthy democracy.

Over the four Lok Sabha terms covered by the dataset, 91.6% of questions were asked by men while 8.4% of questions were raised by women. The average gender ratio of men to women was $8.3\!\!:\!\!1$ over the same four terms. Thus the skewed gender ratio correlates with the distribution of questions. It is noteworthy here that the Women’s Reservation Bill proposing the reservation of one-third of the seats in the Lok Sabha for women has been pending since 2010 [Marwah, 2019], thus, allowing the bias to continue. As regards the caste distribution, 80.6% of questions were raised by MPs from the general caste while the rest come from the reserved categories. Note that 24.03% of Lok Sabha seats are reserved for the reserved categories while the rest belong to the general caste. Figure 3 shows the number of questions on gender and caste-related issues asked in the Parliament; Appendix 6.5. lists the keywords we used for this analysis.

We have selected two topics, ‘banking’ and ‘pension reforms’, and plotted the probability of a few selected tokens in them as a function of time in Figure 6. In the topic ‘banking’ shown in Figure 6(a), we find that the word ‘credit_card’ peaks in 2007. Our analysis shows that most of the questions around this time are related to the growing credit card frauds in India and the sudden rise in credit card interest rates by some banks, coinciding with reports in mass media. Another aspect of credit card-related discussion is Kisan Credit Card (KCC) – a low-interest credit card for farmers, which was introduced in 1998 and significantly improved in 2004. The steady rise in discussion on debit cards correlates with the increasing adoption of debit cards (that avoided the debt trap of credit cards) in India. With demonetization and increased government emphasis on end-to-end digital – as opposed to cash – transactions, terms like ‘atm’, ‘digital_transaction’, and ‘cyber_security’ gain prominence while the popularity of more traditional mediums like ‘cheque’ reduces.

Figure 6(b) shows the topic of pension reforms. In late 2003, the Government of India notified that it was abolishing the then-existing government-funded pension system for all its new employees and that they would come under the National Pension System (NPS) to be administered through the new Interim Pension Fund Regulatory and Development Authority (PFRDA). NPS enabled subscribers to make planned savings for post-retirement income. NPS was extended to all Indian citizens in 2009. Being a monumental change, NPS provoked a number of questions that peaked around 2011; MPs wanted to know the details of the scheme, including its performance, implementation challenges, extension to unorganized sectors, and even its security. The government introduced the Voluntary Retirement Scheme (VRS) in nationalized banks in the early 2000s to reduce the financial load on the public exchequer. Thousands of employees across various organizations accepted VRS in 2000-2001. Given the high unemployment rate in the country, VRS generated a lot of panic among people and pointed questions in the Parliament on the government’s future plans for its workforce. The other visible terms ‘apy’, ‘dbt’, and ‘jan_dhan’ refer to recent financial inclusion programs.

Observe the peak of the term ‘public_private’ around 2010 when the Railways ministry introduced a new model for a public-private partnership to modernize the Indian Railways. Indeed, there have always been questions and panic in the Lok Sabha on the public-private models as privatization of the economy could increase fares, job loss, and casualization of labor [Makhija, 2006, Reddy, 2019]. A related term ‘joint_venture’ was a part of many discussions. While early uses of it (in the 2000s) focused on joint ventures of Indian Railways with other public sector companies, the recent focus (since 2015) has been on the increasing role of private players. Indeed similar exchanges occurred between the MPs and the Civil Aviation Ministry on the privatization of airlines. Terms like ‘rail_vikas’ and ‘irsdc’ refer to companies owned by Indian Railways and entrusted with the maintenance of Railways. The rise in freight volumes led to the ideation of Dedicated Freight Corridors (DFC) in 2005 and generated a number of questions (‘dedicated_freight’) related to their cost, progress, and expansion. Discussions on the introduction of bullet trains have been present for a long time but they gathered momentum when a vision document was tabled by the government in December, 2009 and the construction of the Mumbai-Ahmedabad high-speed rail corridor started in 2017.

Here, we have chosen the topic ‘Communicable Diseases’. Figure 8 shows the evolution of word probabilities in this topic. Clearly, we find an increasing focus on the word ‘virus’ because India has been repeatedly hit by the Swine influenza virus (such as H1N1), including in 2009 when swine-flu turned into a pandemic and in 2014-15 [Kshatriya et al., 2018]. MPs in the Lok Sabha enquired about the number of cases, test administration, government interventions, vaccination drives, the role of WHO, and the effort to develop indigenous vaccines. Another major disease in India has been malaria but a response in Lok Sabha informs us that malarial death reduced steadily over the years, and that is attested to by the steady decline in its focus in the Parliament. India recorded thousands of deaths due to Acute Encephalitis Syndrome (AES) in 2008-14 [Ghosh and Basu, 2016] and Japanese Encephalitis in 2005-11 [Adhya et al., 2013]. During these unfortunate occurrences, the Lok Sabha witnessed intense discussions on diseases.

We have removed the punctuation from the dataset, then lowercased and lemmatized the words. We have also removed the stopwords and filtered out the remaining words having document frequency lower than $0.001$ and higher than $0.95$. We have only kept the words that have at least 3 characters. Finally, we removed the documents with less than 3 words in them. After preprocessing, we created bigrams to better capture word co-occurrence statistics.

We used the following hyperparameters to run LDA, LDAseq, and D-ETM.

1. 1.

   LDA [Blei et al., 2003]: We use the Gensim implementation of the LDA model. To enable reproducibility, we use a fixed random seed, i.e., set the random_state $=2021$. We set passes to $20$ and use the default values for the remaining hyperparameters.

2. 2.

   LDAseq [Blei and Lafferty, 2006]: We use the Gensim implementation. We set the random_state value as $2021$ and passes as $20$. For the rest of the hyperparameters, we use the defaults.

3. 3.

   D-ETM [Dieng et al., 2019]: We use the original implementation¹¹1 https://github.com/adjidieng/DETM with batch_size $=64$ and epochs $=100$, and keep the default values for the rest of the hyperparameters.

To generate keywords for the analysis of gender and caste-related discussions, we have applied SkipGram, which is one of the models used in the neural network-based word2vec algorithm to generate word embeddings [Mikolov et al., 2013]. We have run SkipGram on the entire TCPD-IPD dataset and taken the top twenty neighbors (based on the cosine similarity of the generated word vectors) of each of the keywords ‘gender’ and ‘caste’. The words are shown in Table 4. Then, we counted the documents that contain those words. We have ignored the words shown in italics in the table as they introduced many irrelevant documents into the count.

Words in the main text that are difficult to understand outside the Indian context are explained below.

1. 1.

   ‘npci’: National Payments Corporation of India, created by the Reserve Bank of India under the Ministry of Finance, to enable digital payments and settlement systems in India.

2. 2.

   ‘nabard’: National Bank for Agriculture and Rural Development, operating under the Ministry of Finance. It regulates the institutions that supply financial help to rural society.

3. 3.

   Kisan Credit Card (KCC): Farmers’ Credit Card.

4. 4.

   ‘apy’: Atal Pension Yojana. ‘Yojana’ means scheme.

5. 5.

   ‘dbt’: Direct Benefit Transfer.

6. 6.

   ‘jan_dhan’: Pradhan Mantri Jan Dhan Yojana. Translates to ‘Prime Minister’s People’s Wealth Scheme’.

7. 7.

   ‘rail_vikas’: Railway Vikas Nigam Limited is owned by the Ministry of Railways involved in building rail infrastructure.

8. 8.

   ‘irsdc’: Indian Railway Station Development Corporation.

9. 9.

   ‘scs’, ‘obc’, ‘obcs’, ‘scs_sts’, ‘scheduled_tribe’, ‘scheduled_caste’: These words denote historically disadvantaged communities in India. Scheduled Castes, Scheduled Tribes, and Other Backward Classes are abbreviated as SC, ST, and OBC, respectively.