But some of it was reasonable enough:
Yes, following the link would’ve revealed that it applied to less than one-quarter of all responses, but the issue is worth addressing.
If we wanted the results of the poll to put the FBI/DOJ/Deep State in as unfavorable a light as possible, the graph would’ve looked something like this:
Just half of Clinton voters think the FBI is acting impartially, while majorities from every other group do not think they are doing so! Looks positively catastrophic for the FBI, as opposed to just bad, as is actually the case.
In the same way reporting that polling on the eve of the 2016 presidential election showed Hillary Clinton’s support at 26% would have been technically true, since she garnered just under 66 million votes out of an adult population of 250 million, presenting it like this is intentionally misleading. There’s a reason serious polling outfits restrict responses to likely voters, or preface questions to registered voters with “if you were voting today, …”
Presenting the three different responses across 11 different demographic groups would leave us with 33 bars in the graph. It would be more obfuscating and overwhelming than clarifying.
What I’m always after in these exercises is the expression of general sentiment in a clear, concise way that is easy to comprehend. The source data is, except in the rare case where something is paywalled, directly linked to and freely accessible, and notes on presentation–in this case, that “unsure” responses were excluded–made clear.
In this case, the figures are from Reuters-Ipsos raw survey data, not from a write up on the results. Parenthetically, the poll was only administered for a week, no public release was made (so far as I’m aware), and it never showed up on the interactive polling explorer site’s main page. It’s almost as if this little doozy is one the bosses weren’t too keen on getting out!