“… The Green-Feminine stereotype and its effect on sustainable consumption”


Updated, 2016-11-28

I heard about this study earlier this year, and queued it up for a careful examination. I got to that today. The article is:

A R Brough, J E B Wilkie, J Ma, M S Isaac, D Gal, “Is Eco-Friendly unmanly? The Green-Feminine stereotype and its effect on sustainable consumption”, Journal of Consumer Research, December 2016, 43(4) 567-582.

It’s not a study I would draw deep conclusions from, and I find their generalizations unwarranted.

Sample smudges:

  • Based upon 7 studies. One study had 127 university students, using a hypothesis drawn from the literature. 52% are male. A second had “… 194 students (45.9% male; mean age = 23.05) simultaneously recruited from two private universities to participate in an online survey.” (Emphasis added.) A third had “… 131 individuals (58.0% male; mean age = 35.21) recruited on Mechanical Turk to participate in an online study session.” A fourth had 403 “… American men (mean age = 32.68) were recruited from Mechanical Turk to participate in the study.” (Note that on the fourth study, the scholars qualified the results with “Although only males had been recruited to participate in this study, fourteen participants reported their gender as female and were therefore excluded. This left 389 participants for the analysis.”) A fifth had 472 “… participants (49.4% male; mean age = 35.29 recruited from Mechanical Turk completed the study.” There were also studies 6A and 6B, whose descriptions add no insight.
  • For the interpretation of the first study, “There was no difference in IAT D-score by participant gender; F(1, 57) = .11, p = .74, ηp2 < .01, suggesting that both men and women cognitively associate the concepts of greenness and femininity.'' I need not say more. They are erroneously interpreting a high p-value as meaning the null is confirmed. Later “… the target’s gender and environmental behavior remained significant predictors of femininity (p’s .37).”
  • The litany of woes continues in later summaries: “Because the two dependent measures were highly correlated (α = .87), we averaged them to form a composite evaluation measure.”

The scholars draw overly strong conclusions from the limited data in hand. For example, they write:

… [W]e provide the first experimental evidence of the implicit cognitive association between the concepts of greenness and femininity (study 1), and show that this association can affect both social judgments (study 2) and self-perception (study 3) among both men and women. Focusing on the downstream consequences of this green-feminine stereotype, studies 4-6 suggest that as a result of gender identity maintenance, gender cues (e.g., those that threaten or affirm a consumer’s gender identity or that influence a brand’s gender associations) are more likely to affect men’s (vs. women’s) preferences for green products and willingness to engage in green behaviors.

Further, they have the audacity to claim “More generally, our findings also add to a growing body of research pointing to a link between identity and consumers’ tendency to engage in sustainable behavior.” What “growing body of research?” A correlation-only-based study like

J A Lee, S J S Holden, “Understanding the determinants of environmentally conscious behavior,” Psychology and Marketing, 1999, 16(5), 373-92?

Apart from the sampling issues (Mechanical Turk? Really?), the wholesale neglect of repeated uses of the same population for successive tests with no corrections (even if this passes a `smell test’ in their field, and, obviously, satisfies their peer reviewers), and the failure to estimate in-sample versus out-of-sample effects through some kind of bootstrap or cross-validation means that, for all we know, these conclusions are limited to the samples the scholars took. Since Turk was used in most of the tests, it could not have hurt to repeat the same study with another draw for each from the general population, or seeing how much their p-values varied with matched subset of the samples they had. In the very first study, the scholars reported a p-value less than 0.001. That is extraordinary with a sample size of only 127.

About hypergeometric

See http://www.linkedin.com/in/deepdevelopment/ and http://667-per-cm.net
This entry was posted in Frequentist, statistics. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s