Paint by numbers

We observed what can go wrong when journalists who lack an understanding of statistical inference make claims about data in How Mail Today got its analysis of communal violence exactly wrong. This tendency became clear once again from an Apr 26 article in The Hindu titled Gujarat behind West Bengal in new factory jobs, as the following analysis shows:

Guest post by Philip K. Oldenburg, Research Scholar, South Asia Institute, Columbia University 

Perhaps the most common error the media make when presenting data is to use the “absolute” numbers on their own, rather than adjusting them appropriately. For example, growth in income should always be “real,” adjusted to eliminate the effect of inflation. In a piece headlined “Gujarat behind West Bengal in new factory jobs; National Sample Survey data poses a challenge to ‘Gujarat growth model’,” published in The Hindu recently (April 26, 2014), we find a graph showing that there were 24 lakh jobs created in the manufacturing sector between 2004 and 2011 in West Bengal, versus 14.9 lakh jobs in Gujarat — some 60% more, making the state a “distant second.”

If one adjusts the figures to take into account the differing populations of the various states, the graph would be much less dramatic, and the ranking changes (excluding states with negative job creation):

Jobs created per 10,000 population

Screen Shot 2014-04-27 at 11.07.50 am

West Bengal is now only 8.5% “better” than Gujarat. (Ideally, one would “normalize” the absolute numbers in this case by using some other denominator that would adjust for size of the state: perhaps its adult worker population, or something else.) That adjustment makes the analysis in the rest of the report irrelevant, in my opinion.

In the best of all possible worlds, any reporter assigned to do a story that rests on quantitative data should have had at least a course in elementary statistics, and every editor (equipped similarly) would catch errors in data presentation and analysis with about the same failure rate as now exists for typographical errors.

Deflating expectations

Every time the price of onions rises, pundits remind us (here and here) of what happened in Delhi in 1998, when a spike in the price of onions is believed to have contributed to the state election defeat of the Bharatiya Janata Party (BJP). Inflation does appear to matter greatly to Indian voters: “price rise” was the single most important issue for voters in the February 2014 CSDS tracker poll (PDF), followed by “development”, “corruption” and “economic growth”.

Some blame the United Progressive Alliance (UPA) for feeding food inflation via frequent increases in the minimum support prices (MSP) for cereals like rice and wheat in its quest for rural votes. This may therefore have come as a bit of a surprise:

Modi’s declaration is a timely reminder of just how far we are from India’s “Thatcher moment“. India’s political realities do not permit a mass party like the BJP to ignore key constituencies, and this will continue to shape the economic policies of a future BJP-led government.

That said, inflation is a hot political issue and the BJP’s election manifesto claims that the National Democratic Alliance (NDA) did much more to control inflation than did the UPA:

The BJP-led NDA Government’s record of holding the prices is a demonstration of our commitment to break the vicious cycle of high inflation and high interest rates.

Tellingly, the Indian National Congress manifesto has no mention of either inflation or “price rise”.

How to judge? Just as India’s economic growth is linked to the global economy through trade and investment flows (see How bad is the UPA’s economic record?), the prices of Indian goods and services — particularly those that are tradable — are tied to global prices. I therefore compare India’s annual GDP deflator since 1998 with that of developing countries (using World Bank data). The GDP deflator is a better measure than an inflation index (like the Consumer Price Index) because it does not limit its measurements to a fixed basket of goods and services but one that evolves with the economy.

And this is what we find: the NDA clearly was more successful at controlling prices than the UPA. In absolute terms, inflation rose from 4.5% during the NDA period (1998-2004) to 6.2% during UPA1 (2005-09), and further to 8.2% during UPA2 (2010-12).

Screen Shot 2014-04-19 at 4.46.06 pm

But how much of this can be blamed on poor economic management? That’s where the global comparison becomes relevant, and the UPA2 in particular performs poorly here. Under the NDA and UPA1, India’s inflation rate stayed below that of the developing world (although the differential decreased during UPA1). But under UPA2, Indian inflation soared to 2.5 percentage points above the developing country average. This can’t simply be explained away by high oil prices and inflation in food commodities (e.g. wheat and rice) because developing countries were also affected by those.

This World Bank chart shows how India’s inflation rate (in %) diverged quite sharply from other countries from 2009 onwards:

Screen Shot 2014-04-19 at 1.14.22 pm

So why did this separation occur? Plausible reasons include:

  • Fiscal deficit. The UPA sharply increased government spending in 2008-09, in the run up to the 2009 elections as well as to stave off the effects of the global crisis, but found it hard to reverse course and reduce the deficit subsequently for political reasons. Finance Minister P Chidambaram even blamed his predecessor, now President Pranab Mukherjee, last year for this inflationary surge.
  • Food prices. Food inflation has been a big driver of overall inflation in recent years. As their incomes have risen, Indians have consumed greater quantities of proteins such as milk, eggs, meat and fish, and the prices of these commodities have accordingly spiralled.
  • Minimum support prices. Government policies to ensure that farmers receive higher prices for cereals (e.g, rice and wheat) and pulses have also contributed to food inflation.
  • Subsidy reform. Ironically, price increases in petrol, diesel, non-nitrogen fertilisers and electricity aimed at reducing the deficit have helped increase inflation directly, as well as indirectly by making it more expensive to grow agricultural produce.
  • Rural wage increases. The wages of agricultural labourers have gone up rapidly, both because of the rural employment guarantee act and because of higher wages in sectors like construction that provide alternative employment for this section of workers.

Not everyone agrees that persistent inflation is a completely bad thing. Harish Damodaran argues that inflation in India is a consequence of the economic empowerment of lower-income groups.

For now, though, it’s the UPA that seems to be paying a political price for elevated inflation and slow economic growth.

Postscript

Some analysts like Yogendra Yadav do not think that the concept of “price rise” as presented in opinion polls is reducible to inflation. As he wrote in 2009:

In India ‘price rise’ is a way of talking about the lack of purchasing power or insufficient income rather than what economists call ‘inflation.’

If so, then real wage growth might better capture how the state of the economy will affect voting behaviour. But that’s a different study.

Is AAP India’s most criminal party?

With candidates declared for six of the nine rounds of voting in this never-ending general election, a partial analysis can now be done of how many stand accused of criminal activity and in what proportion parties are awarding tickets to such folks (data from National Election Watch, also see preliminary analysis here).

The chart below ranks parties by the proportion of candidates that have serious criminal charges against them:

Screen Shot 2014-04-22 at 1.45.48 pm

As before, parties from Bihar and Maharashtra have the highest percentage of Lok Sabha candidates that face serious criminal charges. In 2009, the Indian National Congress (INC) and Bharatiya Janata Party (BJP) were comparable; the BJP is doing a little worse so far in the 2014 election with 17% of its candidates facing serious charges compared with 13% of the INC’s.  The Bahujan Samaj Party (BSP) is performing better this time round: it has fewer accused candidates than the BJP, INC or the Samajwadi Party (SP). The Aam Aadmi Party (AAP), as might be expected, looks pretty good here.

What about levels of criminality? To calculate the intensity of the alleged criminality of each party’s candidates, we count the number of candidates with more than one serious charge, add up the number of charges and calculate the total as a percentage of the total candidates per party. And here is what we find:

Screen Shot 2014-04-21 at 11.40.40 am

Shock and horror! AAP turns out, by far, to have the highest levels of alleged criminality of all of India’s parties. The Shiv Sena and Nationalist Congress Party come a poor second and third respectively.

As it happens, AAP’s ranking is driven by two statistical outliers: Anti-nuclear activists SP Udayakumar and M Pushparayan respectively account for 382 and 380 of the 829 serious charges that AAP candidates face. The next such candidate is Trinamool’s Kameshwar Baitha with “only” 48 charges, suggesting that the numbers for Udayakumar and Pushparayan are unusual. After all, is Udayakumar really 48 times more “criminal” than Sri Ram Sene leader Pramod Mutalik who faces eight serious charges?

To limit the effect of such outliers, let’s give a score of 10 to all candidates who were had more than nine serious criminal charges. The result is:

Screen Shot 2014-04-22 at 1.46.24 pm

The AAP now drops in the rankings and ends up with about the same intensity of criminal charges as the Congress Party, although it still does worse than the BSP and Biju Janata Dal. The “criminality” of the BJP slate of candidates is more than double that of the INC and AAP, up from being 28% greater than the INC in 2009. The Congress score in 2014 is about the same as it was in 2009, but the BJP’s is sharply higher this time.

As before, the BSP does well in this ranking, although its score is not strictly comparable with the SP’s because the latter is focused much more on Uttar Pradesh and has only two-fifth as many candidates. Communist Party of India (Marxist) supporters can breathe a sigh of relief: it moves from the top of the list in 2009 to somewhere in the middle.

Finally the overall incidence of serious criminal charges is higher (so far) in 2014: 302 of the 533 candidates facing serious criminal charges (the 57th percentile) have more than one charge against them compared with 355 of 1,114 (the 32nd percentile) in 2009.

Remember however that this is still an incomplete list: we await candidate data for the final rounds of voting.

Running to stand still

The Bharatiya Janata Party-led National Democratic Alliance (NDA) continues to advance in the opinion polls.  The NDTV-Hansa survey projects a majority for the alliance if elections were held in March:

The CNN-IBN-CSDS-The Week election tracker is more cautious in its forecast, but has also projected a 20-seat improvement in the NDA’s seats tally over the past two months:

This is a good time to revisit the argument I made two months ago in Why you should ignore opinion poll seat projections. I had written that “Indian public opinion polls are much better at predicting party vote shares than they are at extrapolating to seats won in parliament or state assemblies… because vote shares translate into seats in an unpredictable way under India’s multiparty “first-past-the-post” electoral system.

So what does the latest vote share data tell us?

The NDTV-Hansa poll shows that the NDA has increased its vote share by 4.4 percentage points between February and April 2014, while the Indian National Congress-led United Progressive Alliance (UPA) has slipped 1.2 percentage points:

The CNN-IBN-CSDS-The Week poll projects that the NDA vote share has risen one percentage point and the UPA vote share has fallen by a similar amount:

Screen Shot 2014-04-16 at 10.54.42 pm

If we assume a margin of error of 3%, this signifies no real change in the two alliances’ relative positions. In other words, the NDA’s advantage over the UPA has neither grown nor diminished in any concrete sense over the past two months.

No obvious correlation between turnout and anti-incumbency

Voter turnout in the first three rounds of India’s general election has shown a distinct increase by historical standards, particularly in some of the more populous states.

This has given rise to speculation that this is bad news for the incumbent United Progressive Alliance (UPA).

The UPA is clearly trailing in opinion polls, but this does not mean that a high turnout is necessarily bad news for the alliance. The chart below shows no obvious relationship between voter turnout and the level of anti-incumbency in Indian elections.

turnout and vote

That said, the turnout figures for Bihar and Uttar Pradesh are striking, and it will be worth watching whether they persist in subsequent rounds of voting in those states. This could be good news for the NDA if the increases are being driven by younger voters who — opinion polls show — are leaning towards the Narendra Modi-led alliance.

But if they are being driven by women (as occurred in the 2013 state elections) this could offer some relief to the UPA. Although the UPA performed poorly in those elections, both the CNN IBN-CSDS-Lokniti-The Week (see tables 3a and 3c) and NDTV-Hansa Research Group polls show that the female vote nationwide is divided equally between it and the NDA, while the male vote is decisively tilting in favour of the NDA.