I've looked through the paper and this bit showed some inadequacy of the research process. I may agree that branding campaigns are really hard to track and the study conclusions may apply to them. Running branding campaigns without a clear conversion event is dumb though.
On the other side of the spectrum performance-driven(cost per action, cost per lead, cost per sale) campaigns that include monitoring of all parts of customer journey* offer precise ROI data that doesn't need inference.
This study is limited in its scope and doesn't take into consideration the best methods for tracking ROI for digital ads, concentrating on still popular but outdated method of branded advertising.
*Stimulus-Visit-Lead-Repeat visit-Sale-Repeat sale-Recommendation
Branding is about raising awareness and building positive connection with a company image, not about direct selling. If you have a clear conversion event it isn't branding, it is direct response advertising.
For example, a visitor stayed more than 60 seconds on the about us page or a visitor watched a product explanation video, or number of product searches in Google or Bing increased.
You may argue semantics, though.
2) The conclusion of this paper is daunting. It sounds like it's impossible to reliably know if a campaign was -10/+10% ROI. Only if the campaign was =50% was it then just 'difficult' and not impossible.
Very cool stuff Randall and Justin.