Now that opinion polling has become a growth industry, there is the spectacle of cowboys rushing in where professionals fear to tread. With little by way of expert assistance to help distinguish between honest exercises and completely spurious predictions, there is understandable scepticism in the political class to treat opinion polls very seriously.

Despite parties spending many millions of rupees commissioning polls, the tendency is to gloat over favourable projections, rubbish bad news and, ultimately, rely on instinct and intuition. This is why Arun Nehru's armchair psephology has such a large fan following among politicians.

One of the factors that make opinion polls marginal to political calculations is the relative lack of expertise in evaluating the results. There is a spurious belief that the size of the sample is crucial in accurately gauging voter preference. Since political parties invariably ask how a sample of 16,000 or so voters can be a true indication, the tendency is to increase the sample size dramatically. These days, pollsters have taken to flaunting large sample sizes to argue that their poll is more reliable than their competition.

This is a flawed assumption. The reliability of a national poll does depend on a minimum sample size but the accuracy depends on whether or not the sample is random. If the pollsters diligently follow scientific statistical norms, the chances of the poll being accurate are higher. A 16,000 sample is as likely to give an accurate snapshot as a 60,000 sample if the random principle is honestly observed. Conversely, a one lakh sample drawn from interactions at bus stops and tea shops --the familiar journalistic technique -- is likely to go all wrong. In Rajasthan, for example, the pollsters went all wrong for the assembly 2003 polls because adequate responses from women voters weren't taken into account.

It naturally follows that the analyst must not distort the findings by injecting a subjective or ideological bias. This, unfortunately, is easier said than done. I have only too frequently come across pollsters who attach greater weight to the respondents from particular social classes. This is why, for example, the A C Nielson findings were differently interpreted by NDTV and The Asian Age. During the 2002 Gujarat assembly poll, one charlatan masquerading as pollster predicted a neck-and-neck race because he expediently lumped all the undecided respondents into the column for the Congress.

Second, a cardinal principle of opinion polls is that larger the area of concern the more the chances of the outcome being accurate. An all-India poll is likely to yield more accurate results than a state-wise or a constituency survey. In the pre-news channel days, pollsters used to break up India into four zones. Nowadays, with the emphasis on television ratings, there is a tendency to forecast results from the states. This is misleading.

The results of two polls released last week suggest that there has been a dilution of the NDA's lead over the Congress-led alliance in the span of one month. The India Today-ORG-Marg poll, which I rate as by far the most rigorous exercise, suggests that the NDA tally is likely to be 282 seats, down from 335 in January. The NDTV-AC Nielson poll also says that the NDA will barely reach the majority mark of 272.

The facile conclusion of the instant pundits is that in the past two months India has lost a bit of the feel good euphoria and that this is dragging the NDA tally downwards. There is also a suggestion that the assault on Sonia Gandhi's leadership credentials is generating a backlash.

Neither of these theories is backed by the evidence. The feel good, while still not uniform, is still reflected in the staggering approval ratings for Atal Bihari Vajpayee. At the same time, Sonia's Italian origin remains her foremost handicap.

The point which the pollsters haven't taken adequate note of is the fact that the real difference between the January and April surveys is that the second was conducted after the candidates had been announced. To my mind, this is the foremost factor in dragging the NDA down.

The BJP, despite conducting two exhaustive constituency-level surveys, was extremely wary of changing sitting MPs, including those whose constituency records are not terribly good. The party, it seems, lacked the political will to change second and third term MPs because they enjoyed the backing of one or the other national leader. According to some participants, the party's parliamentary board meetings were marked by a complete unwillingness to take tough decision because that would offend or another faction.

Therefore, while there is no significant anti-incumbency for the Vajpayee government, the NDA has been dragged down by at least two percent by the anti-incumbency for the sitting MP. Indeed, there are reasons to believe that the candidate selection of the Congress was better than that of the BJP and that this is yielding limited results.

I observed this most acutely in Uttar Pradesh, Bihar and Jharkhand, three states which are quite crucial to the BJP. The BJP campaign, it seems is strongest in places where the candidate selection has been right. In large cities like Agra, Kanpur and Mathura, where its candidates lack local clout, the campaign is faltering. Likewise in Jharkhand, the BJP has been able to offset some of the strong anti-incumbency against the state government in places like Koderma and Hazaribagh, where it has fielded strong candidates. This is certainly not the case in constituencies like Dhanbad.

In the first phase of polling, the BJP will have to do its utmost to ensure that a high-voltage national campaign subsumes the anti-incumbency of its sitting MPs. The Congress, on the other hand, will have to concentrate all its efforts on local campaigns because its national thrust does not seem to have had an impact.

In the slog overs, what counts are nerve and strategy.