Pollsters also include an error in their figures which is usually around 2-3% mark. When the % being reported is 0.168%, a 2-3% error rate (taking a polling standard) means that the numbers being reported are statistically indistinguishable from zero.
Okay, I'm being a bit harsh because what pollster do and what this study is doing are two completely different things, but to me that 0.168% figure is so low that without some rigorous methodology and long-term history behind it (which applies to medical statistics that can be of a similar order of magnitude), the numbers aren't reliable.
I've just gone to the article that was linked and it appears that this was a study of tweets from June 2014 to December 2015 - either 18 or 19 months, depending on whether the endpoints were included and whether they were whole months. That's actually not a bad baseline, but I'd like to see the level of variability over that time. I'd also like to see this extended out over 5 or 10 years, possibly with trends over that time. For example, that period covers the Supreme Court decision in Obergefell vs Hodges in June 2015. I would suspect that there was a large spike of anti-gay tweets in response to that decision, but how much did that spike affect the overall averages? Sometimes an average can hide the volatility in the underlying data. One or two large spikes in response to one-off events can distort the underlying average.
In short, without some indication of variance (they should be able to calculate standard deviations, given the raw data), it's not possible to judge if the numbers supplied are meaningful.