| There's some rather more fundamental issues with the paper. Let's do a little bit of a dive... 1. The "BSR" is 10 questions on a Likert scale with extraordinarily vague labels. So, what's the difference between "somewhat profound" and "fairly profound"? How confident are you that different populations (eg lib v con) will have similar views on the difference between "somewhat" versus "fairly"? 2. Liberal/conservatism meanwhile is a single question on a Likert scale. 1 to 7, how conservative are you? So, self-image not actual conservatism. And given 109 participants rated themselves on the liberal side versus 46 on the conservative side, it's going to be dominated by "just how extreme do you think your liberalism is?" 3. But best fun of all - On the left, we have
1 = liberal
1 = not at all profound
1 = not at all favourable On the right side of their questions we have
7 = conservative
5 = very profound
5 = very favourable 109 participants were Liberal (less than 4 on lib/con Likert item)
46 participants were Conservative (above 4 on lib/con Likert item) So, just the factor of "how much do you like to tick the extreme boxes on a Likert scale" would give a correlation like the one they get. More likely to pick a 1 than a 2 on a Likert item? You'll rank as both more liberal and less receptive to bullshit then... Like to leave a radio box on the left so you don't feel extreme? That'll register you slightly more conservative and slightly more receptive then... And as the participant pool is 2:1 liberal:conservatism, then that extremeness factor will produce candidate correlations like the ones they get too. (More extreme-tickers are likely to be going for 1s on lib, 1 not profound, 1 not favourable of Republicans, and 5s on favourability of Democrats. Middle-of-the-road tickers are likely to be going 2s for lib, 2s for profound, 2s for Reps, and 4s for Dems they like. Higher score for bullshit receptivity, less liberal, less favourable of Dems, and more favourable (less unfavourable) of Republicans. MeanMundane is bang on the middle (3.1 mean), neatly unaffected by "extremeness" factor, whereas MeanBullshit isn't (2.6), so extremeness will push out the "controlled" correlations neatly too. (Yes, I'm procrastinating, and had a brief back-of-the-envelope poke around their CSV of data...) |