Populism in Power

02 · The measuring stick

Compared to what?

Nobody can rerun the last decade without Modi. The next best thing is a synthetic India: a weighted blend of other countries, with the weights chosen so that the blend tracks India’s actual path over the thirty years before Modi took office, 1984 to 2013. If a blend reproduces India for thirty years, its path after 2014 is a reasonable estimate of where India was headed. The test is then the one Modi set for himself: did his India do better than that?

Income and governance get separate pools of eighteen comparison countries each, chosen by explicit rules on population, income and democracy scores rather than by hand. Significance comes from placebo tests: every comparison country gets the same “treatment,” and we ask whether India’s divergence is larger than theirs. The rules, the lists and the estimators are in the researcher section and the paper.

03 · Maximum governance?

Nine of ten measures of democracy got significantly worse

Ten indicators from the Varieties of Democracy project, each plotted as India minus its synthetic counterpart. A line along zero would mean Modi’s India performed exactly as the no-Modi India would have. The dashed line marks 2013, his last year out of office.

The pattern is the same almost everywhere: a flat line for thirty years, then a drop starting in 2014 that does not reverse. Equality before the law and equal protection improve briefly before falling. Legislative constraints on the executive is the one measure that recovers; it falls sharply and then returns to roughly its counterfactual by 2023. The counterfactuals themselves barely move after 2014; the change is all on India’s side.

In seven of the ten cases India’s divergence is larger than that of every placebo country (a p-value of 0.00). For corruption and equal protection it is the second largest of nineteen (0.05), and for freedom of religion the third largest (0.11).

Modi made fighting corruption a central part of his campaign. Yet by 2023, India scored substantially worse on V-Dem’s corruption index than our synthetic India. Not exactly the result his supporters were promised.

Beyond the indices

The V-Dem indices are not the only evidence. Committee review of bills, investigations of opposition politicians, religious hate crimes, foreign-funding registrations for civil-society groups, and government blocking of online content all moved in the same direction after 2014. We have put that material, with its sources, on a separate page.

04 · The 60-month economy

Was it at least worth it for the growth?

This is the bargain many observers assumed was on offer: give up some democracy, get development. So here is the economic half of the report card, real income per person in India against the India without Modi.

By 2023 the average Indian earns about $1,100 a year less than our estimate of what they would have earned without Modi, roughly 11% of what their income would have been. In five of the ten years India’s shortfall is larger than every placebo we can construct (p = 0.00); in the other five it is beaten by exactly one (p = 0.06).

The statistical evidence is stronger in some years than others. But here’s the important point: Modi didn’t promise to match India’s previous performance. He promised to beat it. We find no evidence that he did. The 60 months came and went, and so did 120.

But what about Covid? Or demonetization? Fair questions. The income gap was already there before Covid, and demonetization was Modi’s own policy. It’s part of his record, not something we should set aside when judging it. And if India’s post-2015 GDP revisions overstate its growth, as several critics argue, our estimate is a floor, not a ceiling.

05 · The elections

So why does he keep winning?

There’s an obvious question here. If India has performed so poorly relative to our estimates, why does Modi keep winning elections? We can’t answer that question with these data. But one possibility is that some groups have benefited from his policies, or feel that their interests and values are better represented, even if the country as a whole has fallen behind. Economic performance isn’t the only thing voters care about. And what benefits the country as a whole isn’t necessarily what matters most to every voter. The paper sketches a simple model along these lines.

06 · The verdict

Which India did we get?

Less democracy, but more prosperity? That might have been one way to defend Modi’s record. But that’s not what we find. Compared with our synthetic India, the real India ended up with weaker governance and lower income. So much for the promised economic miracle.

We are not making a case for the Congress Party. It was not great. But on nine of the ten governance measures we study, and on income, Modi and the BJP did worse than the India they inherited was on course to do.

Achhe din aane wale hain, the campaign promised: good days are coming. The data reply bure din aa gaye (बुरे दिन आ गए): the bad days have come.

07 · The response

What the government said, and what we said back

On 25 August 2026 The Wire reported the paper’s findings. The next day a Senior Advisor at the Ministry of Information and Broadcasting sent the editors a rejoinder headed “Not ethical journalism.” The Wire published it with our reply. The full exchange is here.

The rejoinder made three claims: that the donor pool was “cherry-picked” and a “complete” pool would flip the result; that V-Dem is “an opaque index of subjective judgements by anonymous experts”; and that a long list of achievements since 2014 shows the economy has been transformed.

Our reply, in brief. The donor pool was chosen once, before we saw any result, and the inventor of the method is explicit that adding poor matches to a pool is “a recipe for bias,” not a robustness check. V-Dem is the standard governance data in political science and the opposite of opaque: it publishes its procedures, its coder-level data and its uncertainty estimates, and it scores our donor countries by the same process it scores India. And nobody disputes that India has grown or built things since 2014; the question is whether it did better than the thirty years before would have predicted, and a list of accomplishments does not answer that. One footnote: Freedom House scores India on “free private expression of personal views without fear of surveillance or retribution” at 4 of 4 in 2012 and 2 of 4 in 2025. Given the tone of the exchange, we can see why.

The critique was aimed at the August version of the paper, with hand-chosen pools of fourteen and twelve donors. The version on this page uses rule-based pools of eighteen, and the results are the same.

08 · For researchers

How we tested it

Open any of these for the details, or go to the paper and the replication files.

Is the model just unstable out of sample? (In-time placebo)

Pretend the treatment began in 2004, when Manmohan Singh became Prime Minister, end the sample in 2013, and rerun everything. If the method were manufacturing gaps, it would manufacture one here. It does not: the yearly p-values run from 0.22 to 0.94 and the joint p-value is 0.33. India under a fake treatment looks like a typical member of the donor pool.

Cherry-picked predictors? (All-lags specification)

Researchers can tilt synthetic control results through their choice of matching variables. Following the advice of that literature, we re-estimate using every pre-treatment year of income and nothing else. The effect is smaller, −$593 rather than −$1,095 by 2023, but has the same sign and the same pattern of significance.

“But 36% Ethiopia?” (Leave-one-out)

Against our better judgment, we also report a leave-one-out exercise: the all-lags model rerun without Bangladesh, then without China, then without Ethiopia. The three countries are in the pool because they meet the same rule as everyone else, and they get weight because their income paths help reproduce India’s. Dropping Bangladesh or China barely moves the result (2023 effects of −$461 and −$498). Dropping Ethiopia makes the pre-treatment fit worse and the estimated shortfall larger, −$851 in 2023, the most extreme in the placebo distribution. Objecting to Ethiopia makes Modi look worse, not better.