I knew it was bad but that bad??!!!! (Mobile phones was part of the problem but far from all. People just do not want to answer any longer.)
Anyway, from that figure one can deduct that if candidate A and B both have 50% of the likely voters, but for some reason candidate A's voters are only 7% likely to answer a poll, whereas candidate B's voters are 9% likely to answer (for all kinds of different reasons) then the result of that poll, assuming otherwise perfect sampling would be 44/56.
In other words unless the pollsters can make certain that their sample is representative of the mass of people asked that declined to answer then the poll is completely worthless - even if it has been weighted for gender, party affiliation, age, income etc etc.
In the olden days when polling companies accepted (but just) a response rate of 80%, then if candidate A and B had response rates 75% and 85% respectively then the outcome would have been 47/53. Thus, the result was much less dependent on how willing the different voters were to answer.
deduct = deduce
Sorry.