Non-scientists are often frustrated by the inability of researchers to give a simple, straightforward answer to a question such as, 'Is red wine good for my health?' They don't like the way that scientists always seem to qualify their conclusions with terms such as 'may', 'possibly', 'indicates' and 'suggests'. Journalists are no better: these qualifiers don't look good in headlines, and the uncertainty that scientists are used to living with in their research is commonly jettisoned when it comes to media coverage of science stories. But science rarely progresses in headline-grabbing 'breakthroughs'. Instead it advances by the steady accumulation of evidence, until a critical mass of data is reached and the consensus changes. And this is how it is with research into the health effects of alcohol.
How do you test scientifically whether drinking wine is beneficial or not? Essentially, you need to take different populations who differ only in their wine consumption, and since the diseases that alcohol may be protecting against develop over many years, these populations must be tracked over a number of years -- ideally, for decades. Added to this, in order to get robust statistics, you'll also need a large population. In practice, meeting all these conditions is just about impossible.
First of all, there are what are known in the trade as 'confounders'. Wine consumption is related to other variables, such as social class, that are themselves related to health factors. Age, sex, race, smoking, education and ethnic background are all related to alcohol consumption and are predictors of chronic disease. These confounding factors need to be adequately controlled for in any study, and in the better studies they usually are taken into account.
Second, there is the thorny problem of tracking how much people are drinking. Unless you monitor your subjects in a 'Big brother'-like fashion, you must rely on their self-reported drinking habits for your data. And drinking is almost always under-reported. I've heard it said that if you extrapolate how much people say they are drinking out to the whole population, and then look at the records kept by Customs and Excise of exactly how much alcohol is sold, there is a difference of around a factor of two. Either people aren't telling the whole truth, or there is an awful lot of booze being poured down the drain!
A third difficulty in these studies is that drinking patterns are not only variable, they change over time, while risks and benefits accrue over decades. This touches on the issue of drinking patterns: intuitively it would seem to be much safer to drink a little each day rather than bingeing at weekends, but few studies factor the patterns of consumption into their design.