When it comes to researching user's screen resolution, I'm a strong believer that vertical positioning matters most when considering key issues such as location of identifying elements like site name and h1-type headings, and elements that satisfy business rules such as ad location. The B&A article, "Blasting the Myth of the Fold" addresses this issue well, and I'm in agreement that "the fold" matters little if the above issues are dealt with. But the issue of browser width is still one that is heavily dependent on audience, and knowledge of what your web analytics tool records about actual users.
When you have a site that has not gone public yet, however, you will have the need to address the issue of resolution in the absence of real data for your site. So where do we turn? We benchmark. Here are a few of my sources for researching this kind of information:
- Start with sites that provide a log analysis service across a broad range of sites and publish their aggregated stats. Example services that provide this data include The Counter, W3Counter, and Net Applications.
- Compare for sanity against individual sites, especially other sites in your industry, or sites you may already publish. W3Schools publishes their stats, for instance, but that audience is heavily skewed towards technical users.
If you have to provide a few sites as a benchmark sources like these may be helpful. Base your browser width decisions on your understanding of who will be accessing your site, just as you would consider user research when deciding what features to provide in your product.
For a bit of information about the estimates you get from aggregated sources, see this Wikipedia entry on Usage of Web Browsers.