# Sustained representation, sustained poverty Hyderabad's old city has returned the same party for a generation, and its households remain the poorest in the district — half the cooking gas, half the internet, a third of the car ownership of the rest of the city, every gap significant at p < 0.0001. This note sets that out, and is careful about what it does and does not establish. One fact does most of the work: the poverty is not explained by missing services. On water, drainage and sanitation the old city matches the rest of Hyderabad district, and on two measures it is marginally ahead. Whatever kept these households poor, it was not the absence of pipes — which strips the easiest alibi off the question of what a generation of the same representation delivered. --- ## The two facts **Representation has been stable and one-sided.** AIMIM leads 26 of the 28 constituency–year rows in this dataset, across six assembly elections from 1999 to 2023, and has held all five seats continuously since 2009. Vote shares run from 30% to 78%. The two exceptions are both Malakpet, and both matter. AIMIM fielded no candidate there in 1999 or 2004; **BJP took the seat with 53.24% and Congress with 52.12%**, both outright majorities. Every seat AIMIM has actually contested, it has led. **Household economic status is far below the rest of the district.** Census 2011 enumerated 108 GHMC wards in Hyderabad district. Comparing the 52 wards in the six old-city mandals against the 56 elsewhere: | Indicator | Old city | Rest of district | Gap | p | |---|---|---|---|---| | LPG/PNG for cooking | 55.9% | 77.1% | **−21.2** | <0.0001 | | Household avails banking | 51.4% | 68.7% | **−17.3** | <0.0001 | | Computer/laptop with internet | 9.1% | 20.0% | **−10.9** | <0.0001 | | Car / jeep / van | 5.9% | 15.7% | **−9.8** | <0.0001 | Ward medians; two-sided Mann–Whitney. Half the cooking-gas penetration, half the internet, a third of the car ownership. ## Schooling and work The asset gaps above are proxies for income. The Census ward tables also carry the things income comes from: literacy and work. Same 108 wards, same method — ward medians, two-sided Mann–Whitney. | Indicator | Old city | Rest of district | Gap | p | |---|---|---|---|---| | Literacy, age 7+ | 82.1% | 86.3% | −4.3 | 0.001 | | Female literacy, age 7+ | 77.9% | 81.7% | −3.8 | 0.03 | | Working, age 7+ | 37.1% | 44.3% | **−7.2** | <0.0001 | | Women working, age 7+ | 17.9% | 25.3% | **−7.3** | <0.0001 | Schooling is behind, but modestly. **Work is far behind**, and the gap survives the first obvious objection: the old city has more children (13.0% of its population is under seven, against 10.7% elsewhere, p < 0.0001), which mechanically lowers a whole-population work rate, so the table uses the age-seven-plus population — and the gap is still seven points, at p = 2×10⁻¹². In Charminar ward 39, 32.8% of the age-7+ population works; in Ameerpet ward 99, 45.7%. Three cautions on reading this. **Female work participation measures norms as well as opportunity** — a low rate reflects who seeks work, not only who can find it, and old-city social norms differ from the rest of the city's. **The wards differ in composition**: the old city has a much lower Scheduled Caste share (3.1% vs 7.7%), younger populations, and different migration histories, none of which is controlled for. And one result runs against the deprivation story: **the male–female literacy gap is smaller in the old city** (6.3 points vs 8.3, p = 0.0008). Old-city women read at a smaller disadvantage to their men than women elsewhere in the district. ### Slum housing: the gap that isn't there GHMC's slum survey — 1,351 notified and non-notified slums with ward numbers, households and population — gives a proxy for the tenure question. Joined to the Census ward populations (99 wards join cleanly; the nine Cantonment and Osmania University wards are outside GHMC's slum survey and are excluded, and the join was validated geographically against slum coordinates), the result is a null: | | Old city | Rest of district | p | |---|---|---|---| | Population in GHMC-listed slums (ward median) | 20.5% | 19.7% | 0.61 | | Aggregate share | 23.4% | 26.6% | — | The old city does not hold more of its people in listed slums than the rest of the district. Its deprivation is not the classic slum form — informal settlements waiting for services. It is poverty in ordinary, serviced housing: Ward 43 in Bahadurpura has 9.7% of its people in listed slums, full water and drainage coverage, and the district's lowest cooking-gas use and lowest female literacy. The city's big slum concentrations sit elsewhere, along the Musi and in the north and west. Caveats: this is an administrative dataset of unstated vintage (GHMC's slum surveys date from the early 2010s), joined to 2011 Census denominators — one ward (Yousufguda, 108) comes out above 100%, which shows the mismatch. "Slum" here means GHMC-listed, which measures official recognition as much as housing conditions, and under-notification in the old city cannot be ruled out from this data. ### The scorecard Of the four constraints the services finding points to — schooling, work, credit, land title — three now have ward-level measures. Schooling: modestly behind. Work: far behind. Tenure (via the slum proxy): at parity. Credit and land title proper cannot be measured at ward level from public data: RBI's credit statistics stop at the district, and title records are not published. Every indicator in this analysis is a rate — a share of households or people — not a total. The comparison is per-capita throughout; what no public dataset provides at ward level is income or consumption itself, which is why household assets stand in for it. Data: [`Data/GHMC-WARD-SLUMS.csv`](Data/GHMC-WARD-SLUMS.csv). Data: [`Data/GHMC-WARD-PCA-2011.csv`](Data/GHMC-WARD-PCA-2011.csv), derived from the ward-level Primary Census Abstract; dictionary in [`Data/README.md`](Data/README.md). ## The services are not the explanation Set those against what a municipal corporation actually builds: | Indicator | Old city | Rest of district | Gap | p | |---|---|---|---|---| | Tap water, treated source | 97.1% | 97.4% | −0.3 | 0.50 | | Water source within premises | 94.8% | 94.5% | +0.3 | 0.10 | | Waste water to closed drainage | 98.4% | 97.5% | **+0.9** | 0.07 | | Bathroom within premises | 98.9% | 99.0% | −0.0 | 0.90 | **Not one is significant. Two favour the old city.** The single worst-served ward in the district on treated water (49.1%) and on closed drainage (49.8%) is *not* in the old city. Ward 43 in Bahadurpura is the clearest illustration: **22.0%** LPG — the lowest in the district — alongside **92.9%** treated tap water and **95.6%** closed drainage. Poor households on a well-served network. So the pipes went in. What did not follow was income. ## Does this still hold? Projecting past 2011 Census 2011 is fifteen years old, and the obvious objection is that it no longer describes the city. The right response is not to hedge but to ask which findings are robust to the gap and which are not. The two families of indicators age very differently, and the difference is arithmetic rather than judgement. **The service finding is ceiling-bounded, so it holds.** In 2011 both groups already sat between 94% and 99% on every municipal measure. You cannot open a meaningful gap from 97 versus 97 when the ceiling is 100, and piped coverage does not run backwards. NFHS-5 (2019–21) puts Hyderabad district at 99.6% for an improved drinking-water source and 99.9% for electricity, consistent with saturation in both halves. Nothing since 2011 could have turned service parity into a service deficit. **The LPG gap has closed.** NFHS-5 records 99.5% of Hyderabad district households using clean cooking fuel in 2019–21. Since the old city is roughly half the district, it cannot sit below about 99% without the other half exceeding 100%. The 21-point gap of 2011 is gone, and Ward 43's 22.0% with it. **The banking gap has probably closed.** Jan Dhan accounts opened from 2014, and by 2019–21, 84% of women in Telangana had a bank account they themselves used. Different unit, different definition from the Census household measure, so this does not settle the question at ward level — but a 17-point household gap surviving that expansion is hard to credit. **Internet and vehicle ownership have no ceiling and no scheme behind them.** These stood at 9.1% and 5.9% in the old city, with no programme driving them towards universality. They are where the economic distance plausibly persists, and nobody has measured them at ward level since. **Which sharpens the argument rather than dissolving it.** Note *what* closed the two gaps that closed. Ujjwala (2016) and Jan Dhan (2014) are centrally sponsored schemes, delivered nationally and uniformly. They did not close the old city's gap because it was represented well or badly; they closed it everywhere at once. Convergence on cooking fuel and bank accounts is therefore not evidence of local development, any more than the 2011 shortfall was evidence of local neglect. Both were set elsewhere. The measures still plausibly diverging are precisely the ones no scheme delivers. ### How firm is this section The ceiling argument needs no new data and is as strong as the 2011 figures. The LPG conclusion rests on one district-level figure plus a bounding argument, and carries three caveats: NFHS "clean fuel" is a broader category than the Census's "LPG/PNG"; the bound assumes ward counts proxy population shares; and NFHS is a sample survey where the Census is a complete enumeration. The NFHS-5 district figures have been confirmed against the official IIPS *District Fact Sheet: Hyderabad* (NFHS-5, 2019–21) — a copy of the original PDF is in [`Docs/NFHS-5-Hyderabad-District-Factsheet.pdf`](Docs/NFHS-5-Hyderabad-District-Factsheet.pdf). Electricity 99.9%, improved drinking water 99.6%, improved sanitation 84.4%, clean cooking fuel 99.5%. Two notes from the factsheet itself: it carries **no NFHS-4 comparison column for this district** (the reorganisation of Telangana's districts made 2015–16 estimates non-comparable), so no district-level trend can be quoted; and it does not report the women's bank-account indicator at district level — the 84.4% figure is from the official Telangana **state** factsheet (urban: 83.0%) and is labelled as such. The banking statement remains deliberately weaker than the LPG one: an inference, not a measurement. ## What this supports, and what it does not **Supported:** the old city has been represented by one party throughout, and is substantially poorer than the rest of its district on every household economic measure available. Those two things are true at once. **Not supported — and the distinction is the whole point:** *That AIMIM caused the poverty.* This is a coincidence of two long-run facts, not an identified effect. Causal inference would need variation in representation, and there is almost none — two contests, in one seat, both before the 2008 delimitation. That is not an identification strategy. See "Making the argument robustly" below for designs that could actually deliver one. *That the old city was denied infrastructure.* The data says the opposite. Anyone arguing under-provision of municipal services will be refuted by the Census. *That an MLA could have changed either.* Urban infrastructure in India runs through the municipal corporation, state departments and centrally sponsored schemes. MLA-LADS is around ₹3 crore a year — small against municipal capital budgets. Across this whole period GHMC and the state were held by TDP, Congress, then TRS/BRS. **AIMIM has not controlled the machinery that builds or funds any of this.** That cuts both ways: it weakens a blame argument and it weakens a credit argument. *That the direction runs one way.* Poor, dense, historically under-invested neighbourhoods produce distinctive political attachments. Representation may be as much consequence as cause. Nothing here separates the two. ## Why the reframing is the stronger argument "The old city has no development" is refutable in one afternoon with this Census table, and it aims at the one thing an MLA has least control over. "A generation of unbroken representation has coincided with a generation of unchanged relative poverty, in a place where the pipes and drains were built" is a harder claim to dismiss, and a more interesting one. It moves the question from civic works — where the old city is doing fine — to economic mobility, where the gap is large, consistent across four independent indicators, and statistically unambiguous. It also raises the question worth actually asking: **what would have had to be different?** If service delivery is level and outcomes are not, the binding constraint is somewhere other than pipes — land title, credit access, formal employment, education. Those are answerable questions, and none of them is settled by a vote-share chart. ## Making the argument robustly The claim people usually want to make from this material is that *AIMIM's wins have done little for the people of these constituencies*. Nothing in this repository establishes that, and two things in it point the other way. This section sets out what would be needed, in rough order of how much work it is against how much it buys. ### Why the current design cannot deliver it The obstacle is not data volume, it is the absence of variation in the treatment. AIMIM leads 26 of 28 contests here; the two exceptions are one seat, before the delimitation, in cycles the party did not enter. Comparing old-city wards to the rest of the district does not isolate representation, because the old city differs in density, housing-stock age, unauthorised construction, land-title status and composition — every one of which predicts economic outcomes independently of who holds the seat. Any estimate from this comparison is the sum of the representation effect and all of that, and there is no way to separate them from within the sample. ### The strongest counter-argument, stated fairly The claim is weakest precisely where AIMIM has had the most power. Municipal service delivery is the domain of the corporation and its corporators, and AIMIM has held a large bloc of GHMC corporator seats across these wards for decades. On exactly those measures the old city is at parity or marginally ahead. Meanwhile household income — where the shortfall is real — is the domain an MLA influences least. So the record, as it stands, is *good* where the party had leverage and *poor* where it had little. Anyone advancing the "did nothing" claim should expect to meet this, and should have an answer before publishing rather than after. ### Designs that would actually identify an effect **1. Close-election comparison — run, August 2026.** The design: pool every AIMIM assembly contest nationally, use the margin of victory as the running variable, and compare constituencies the party barely won against those it barely lost — near the threshold, winning is as good as random. We ran it with TCPD election data (via SHRUG, through 2022) and VIIRS night lights aggregated to post-2008 assembly constituencies. *The sample.* AIMIM contested 324 general assembly elections in the data (1999–2022) and won 38. Within ±5 points of the threshold there are **six contests in the party's entire national history**; within ±10, twenty-one. Restricting to elections with usable lights windows (2014–2019) leaves **13 close contests: 3 bare wins, 10 bare losses**. That is the identification base, all of it. *The result.* Lights growth (Δ log VIIRS, three post-election years against pre-years) was 0.225 log points *lower* after bare AIMIM wins than bare losses — nominally a ~20% slower-growth effect. It does not survive scrutiny. An exact permutation test gives **p = 0.12**. And the comparison is confounded exactly where the night-lights literature warns: the three bare wins — Byculla in Mumbai, Dhule City, Nampally in Hyderabad — are dense urban cores with baseline luminance three times the losses' (32 vs 10 nW), and baseline brightness correlates −0.40 with subsequent growth in this sample. Bright places grow slower in lights for sensor and density reasons that have nothing to do with who won. *What this establishes.* The test is not merely underpowered in principle; it is uninformative in fact, and its diagnostics show why: AIMIM's bare wins occur only in big-city cores, which is precisely where lights stop measuring development. The claim "AIMIM's wins caused slower development" cannot be supported — and neither can its opposite — from any public data now available. Data: [`Data/AIMIM-CONTESTS-1999-2022.csv`](Data/AIMIM-CONTESTS-1999-2022.csv) (every contest with margins) and [`Data/AIMIM-CLOSE-CONTESTS-LIGHTS.csv`](Data/AIMIM-CLOSE-CONTESTS-LIGHTS.csv) (the analysis set). The election data ends in 2022, so the 2023 Telangana and 2024 Maharashtra rounds — which may add close contests — are the natural update. **2. The 2008 delimitation as a natural experiment.** The redraw moved wards between constituencies, so some wards changed which party represented them while their population and housing stock did not. Comparing the 2001-to-2011 trajectory of wards that switched into an AIMIM seat against those that did not is a difference-in- differences with within-ward variation — much closer to the case of interest than any national pooling. The obstacle is the ward-to-constituency crosswalk: GHMC was constituted in 2007 and wards were renumbered, so 2001 and 2011 ward identifiers do not correspond and must be reconciled geographically before anything else can happen. Doing it at mandal level instead is cruder but tractable. **3. Synthetic control on the old city as a unit.** Build a synthetic old city from wards or towns elsewhere in India matched on 2001 characteristics and compare trajectories to 2011. Needs a Census 2001 ward-level donor pool, which is the main cost. **4. Night lights as an annual outcome.** VIIRS from 2012 and DMSP before it give a yearly series, which permits event studies around each election rather than one before-and-after. Two warnings: the elasticity of lights to consumption is much weaker in time series than in cross-section, and a dense, long-electrified metro is close to saturated — so lights are a reasonable outcome for the rural and small-town seats in design 1 and a poor one inside Hyderabad. ### The cheap, direct test of "did nothing" — run The direct test looks at what the representatives actually controlled: the discretionary funds and the legislature. It needs no identification strategy, and the answer turns out to be textured rather than one-sided. **Hyderabad has been an AIMIM Lok Sabha seat continuously since 1984** — Sultan Salahuddin Owaisi to 2004, Asaduddin Owaisi since — which makes the MP fund a clean series. The record: | Term | Fund position | Works | Source | |---|---|---|---| | 16th LS (2014–19) | Recommended ₹24.87 cr; sanctioned ₹16.72 cr; **utilised ₹7.09 cr** (~28% of the ₹25 cr entitlement) | 225 completed | Citizen Matters candidate report card, from MPLADS data | | 17th LS (2019–24) | Released ₹15.51 cr; **spent ₹8.77 cr** (57% of released) | 8 of 69 recommended works completed | Siasat, citing the MPLADS portal | | 18th LS (2024–) | Released ₹9.80 cr; spent ₹0.58 cr | none completed yet (term young) | Siasat, citing the MPLADS portal | **The legislative record points the other way.** PRS Legislative Research, 17th Lok Sabha: attendance 72% (national average 79%), **60 debates** (average 46.7), **366 questions** (average 210), **3 private member's bills** (average 1.5). In the 16th, Citizen Matters reports 82% attendance and **737 questions** against a 293 average, and a Sansad Ratna award for the 15th. Whatever else is true, this is not an absentee representative. **The state-side fund shows the same under-spending — across every party.** Under Telangana's Constituency Development Programme (₹3 crore per MLA per year from 2016–17), Deccan Chronicle reported that in 2017–18 the sixteen Hyderabad-city MLAs together spent under ₹10 crore of ₹36 crore released, and the five who spent below ₹50 lakh included AIMIM's Pasha Quadri (Charminar) and Akbaruddin Owaisi (Chandrayangutta) — **alongside BJP's K. Laxman and G. Kishan Reddy and TRS minister Talasani Srinivas Yadav**. Under-utilisation of constituency funds in Hyderabad is a city-wide, cross-party pattern, not an AIMIM trait. **Reading it honestly.** The MPLADS utilisation numbers are genuinely poor, and they are the strongest single piece of evidence available for the folk claim — discretionary money the MP controlled, left unspent. Two things about them are true at once and should be said together. Leaving the money unspent is a real, directly attributable failure of representation — the only one in this analysis that is the officeholders' own act rather than an outcome they may or may not have influenced. And the money is far too small for its full use to have changed the outcomes measured here: the entire five-year entitlement is about ₹25 crore for a constituency of roughly twenty lakh electors — around ₹250 a head for the whole term, spent on small works — against gaps measured in employment and household income. Unspent funds convict the officeholders of not using their tools; they do not show the tools could have closed the gap. Against that: works are executed by the district authority, not the MP, so completion counts partly measure GHMC and the collectorate; expenditure figures trail recommendations by years as bills post; the 17th-term envelope reflects the national MPLADS suspension of 2020–21; and the cross-party CDP pattern says under-spending is how Hyderabad's fund machinery behaves generally. And the legislative record is well above average on every measure PRS tracks except attendance. "Did nothing" survives contact with this data only in the narrow form: *the discretionary works funds were substantially under-spent* — a statement that indicts the city's MLAs of every party, and its fund administration, as much as its MP. **Still uncollected:** GHMC ward-level works and budget data — the corporation is the right unit for service delivery, and ward-level expenditure would test the parity finding directly rather than by proxy. Telangana publishes no per-constituency CDP utilisation series; the 2017–18 figures above are a single press-reported year. All figures above are as reported at their access dates (August 2026); the MPLADS portal (mplads.mospi.gov.in) revises as bills are uploaded, and none of the press figures has been re-verified against the live portal, which does not expose a static report. ### The version of the claim that survives Stated as "AIMIM has done nothing for the old city", the argument is refutable with one Census table and aimed at the thing an MLA controls least. Stated as follows, it holds: > A generation of unbroken political incorporation has not been accompanied by economic > convergence. Where the old city has caught up — cooking fuel, bank accounts — it was > pulled up by national schemes that reached everywhere at once. Where no scheme > operated, the distance appears to remain. That is a correlation, honestly labelled, about political incorporation failing to convert into economic mobility. It is more interesting than the blame version, it is much harder to dismiss, and it points at answerable questions: land title, credit access, formal employment, schooling. ## Limits of this analysis **Census 2011 is one time point, now fifteen years old.** It pre-dates most of the 2014–2023 period. The word "sustained" on the poverty side is doing more work than one census can carry; establishing it properly needs Census 2001 ward tables, or Census 2027 when it lands. **Mandals are not constituencies.** The six old-city mandals approximate the five assembly seats; the boundaries are not identical. The comparison is between parts of a city, not between electorates. **These are levels, not change.** Levels largely record history. A poor area in 2011 may have been poor in 1951. Detecting whether representation changed a trajectory needs at least two time points, which this does not have. **No controls.** Old-city wards differ from the rest in density, housing-stock age, unauthorised construction and land-title status — all of which predict economic outcomes independently of who represents them. ## Reproducing it Data: [`Data/GHMC-WARD-AMENITIES-2011.csv`](Data/GHMC-WARD-AMENITIES-2011.csv) — 108 GHMC wards, Census 2011 table HH-14 (Houselisting & Housing Census), percentage of households by amenity and asset. Extracted from the Census of India release republished by [OpenCity](https://data.opencity.in/dataset/hyderabad-census-2011-data). `old_city = yes` marks wards in the Bahadurpura, Charminar, Bandlaguda, Saidabad, Asifnagar and Golconda mandals. ```bash python3 - <<'EOF' import csv, statistics as st rows=list(csv.DictReader(open('Data/GHMC-WARD-AMENITIES-2011.csv'))) old =[r for r in rows if r['old_city']=='yes'] rest=[r for r in rows if r['old_city']=='no'] for k in ['tap_water_treated','closed_drainage','lpg_png_cooking','avails_banking']: a=st.median(float(r[k]) for r in old); b=st.median(float(r[k]) for r in rest) print(f"{k:26} old {a:5.1f} rest {b:5.1f} gap {a-b:+5.1f}") EOF ``` Significance is a two-sided Mann–Whitney U on the ward distributions — a rank test, chosen because ward percentages are bounded and skewed rather than normal.