The opinion poll conducted by Markinor in early March this year provides probably the best guide to the possible outcome of the South African elections next week. 3,531 adult respondents were polled across the country. They were then further screened according to whether they were registered, likely, and wanted to vote. The results published are derived from the 2,576 respondents who fitted into this subset of ‘likely voters.'

As reported earlier 64,7% of the ‘likely voters' polled said they planned to vote ANC, down from the 68,3% it polled in October 2008. The Democratic Alliance enjoyed the support of 10,8% of respondents (down from 12,3%), and the Inkatha Freedom Party 2,7% (down from 3,6%.) The Congress of the People, launched after the previous poll was conducted, drew the support of 8,9% of ‘likely voters'. 9,1% of respondents were unsure of who to vote for, spoilt their ballots, or refused to answer the voting intention question.

According to the poll 78,7% of ‘likely voters' are black, 11,5% white, 7,8% Coloured, and 2% Indian. The ANC enjoyed the support of 78,8% of black respondents, COPE 6,8%, the IFP 3,4%, and the DA a miniscule 0,8%. The DA meanwhile enjoyed the support of 59,8% of white respondents, COPE 13,7%, the Freedom Front Plus 4,9%, and the ANC 4,4%.

The poll provides further evidence for the collapse of ANC support among the Coloured population. The DA now enjoys the support of 35,1% of likely Coloured voters, the ANC 22,7% and COPE just behind them with 22,2%. Among likely Indian voters the DA is again the largest party with the support of 29,6% of respondents. The ANC had the support of 16,2% of likely Indian voters, COPE 11,8% and the Minority Front 10,8%. It is striking that COPE enjoys a higher stated support among the minority groups polled than it does in the black majority.

In the Western Cape the DA appears close to attaining an outright victory in the upcoming elections. It enjoyed the support of 43,5% of ‘likely voter' respondents in the province, similar to the 42,8% it polled in October last year, and the 24,4% it polled among registered voters in the equivalent period before the 2004 election (it went on to win 27,1% of the vote.) The ANC received the support of 21,9% of respondents, COPE 16,5%, and the ID 8,2%.

The poll indicates that the ANC is likely to see its support decline in most provinces. However, the ruling party is likely to gain considerable new support in KwaZulu-Natal. The ANC enjoyed the support of 61,4% of ‘likely voter' respondents, the IFP 11,7%, the DA 4,9% and COPE 1,8%. By comparison a Markinor poll conducted shortly before the 2004 elections found that the ANC enjoyed the support of 48,7% of ‘registered voter' respondents and the IFP 24,9%. (The ANC went on to win 47% of the vote and the IFP 36,8%.) Interestingly though the support for all the main parties has fallen since October 2008 with the percentage of uncertain respondents has more than doubled from 7,1% to 16,3%.

COPE has made not insignificant inroads in all provinces barring KZN. Its support among ‘likely voters' polled in the other provinces was as follows: Eastern Cape 15,3%, Gauteng 9,3%, Limpopo 9,9%, Free State 6,6%, North West 6,1%, Mpumalanga 9,1%, and Northern Cape 17,1% (see table 2). In a number of these provinces COPE seems to be hurting the opposition as much as it is the ANC. The number of ‘likely voters' saying they will support the DA has close to halved from October 2008 in KZN, Limpopo, Free State, North West, and Mpumalanga. Markinor also recorded a decline in support for the IFP, ID, UDM, ACDP and PAC between October and March.

The accuracy of this poll will be tested soon enough, when the results of next week's elections are announced. But it is possible to pencil in some possible question marks over Markinor's estimates.

In South Africa opposition supporters have traditionally been more hesitant about declaring a preference (and more uncertain about who to vote for) than supporters of the dominant and perpetually victorious ANC. As a result Markinor polls have tended to understate opposition support and slightly overstate ANC support ahead of past elections (see table here).

Before previous elections Markinor has based its poll results on responses by registered voters. This time around Markinor has further screened respondents through its ‘likely voter' test. Although the company describes this as international best practice, this innovation may exacerbate this pre-existing bias against the opposition in our peculiar local circumstances.

Another related question is whether it has got its racial as well as its provincial weightings right. As noted earlier Markinor estimates that 78,7% of ‘likely voters' are black. If this is an overestimate this would result in an overestimate of ANC support more generally (and visa versa.)

A third question is how accurately Markinor is polling COPE's support. Since 1994 the support bases of the main political parties have been known, and opinion polls have just had to document their relative rise and decline. COPE though is a substantial new player. We have no firm indication yet as to how willing its black supporters are to declare themselves to the fieldworkers of polling companies. As noted previously surveys have consistently underestimated the IFP's real support - even if they have managed to document its relative decline. (The DA's black support, as measured by Markinor, seems surprisingly low.)

Lastly, one thing a poll conducted in March can't measure are the shifts in voter preference since then, or how effective each party will be in getting its supporters to the polls. The ANC has been able to spend (what looks like) hundreds of millions of rands on its campaign. But the news cycle has not necessarily worked to its advantage.

Click here to sign up to receive our free daily headline email newsletter