Millisecond Forums

expressions.d is the D (capital D) measure / score in Greenwald et al. (2003). It is not Cohen's d (lowercase d), although the two are mathematically similar: http://www.jakewestfall.org/misc/D_bound.html

Both the IAT D-score and Cohen's d are in essence measures of effect size of the form (Mean1 - Mean2) / StandardDeviation.
Mathematically, the key difference between the two is how the standard deviation is calculated. IAT D-scores are calculated using the overall standard deviation. Cohen's d ( https://en.wikipedia.org/wiki/Effect_size#Cohen.27s_d ), on the other hand, is calculated using the pooled standard deviation. The consequence is that D-scores and Cohen's d have different boundaries: IAT D-scores vary between -2 and +2, whereas Cohen's d has no finite lower and upper boundaries (d may trail off into infinity in theory).

Thanks so much for your advice Dave!

I wonder if I may ask another question on the Age IAT script. In that script, it is mentioned "Inquisit calculates d scores using the improved scoring algorithm as described in Greenwald et al (2003). Error trials are handled by requiring respondents to correct their responses according to recommendation (p.214)."

Does that mean the expressions.d in the Age IAT summary data file is calculated through Steps 1-12 under "Improved algorithm" column of Table 4 on p214 of Greenwald et al. (2003)?

Or is the expressions.d calculated through Steps 1-3, 8, 9, 11, and 12 under "Improved algorithm" column of Table 4, as well as Steps 4-7, and 10 under "Approximately equivalent alternatives for improved algorithm" column of Table 4?

Thanks again!

> "Does that mean the expressions.d in the Age IAT summary data file is calculated through
> Steps 1-12 under "Improved algorithm" column of Table 4 on p214 of Greenwald et al. (2003)?"

Yes, in essence. But steps 5 and 7 aren't applicable -- the Inquisit scripts calculate D1 ("D with built-in error penalty", cf. Table 3).

Thanks again Dave. I wonder in addition to step 5 and step 7 being not relevant to the calculation of expressions.d, whether "eliminate subjects for whom more than 10% of trials have latency less than 300 ms" in step 2 is also not relevant to the calculation of expressions.d?

That is, is the calculation of expressions.d only based on the following steps under "Improved algorithm" column of Table 4:
step 1;
part of step 2 ("Eliminate trials with latencies > 10,000 ms");
step 3;
step 6;
step 9;
step 10;
step 11;
step 12.

> "eliminate subjects for whom more than 10% of trials have latency less than 300 ms

Those subjects are supposed to be eliminated entirely -- i.e. the recommendation is to not use / analyze their data at all. It is *not* something that factors in to calculating D.

> part of step 2 ("Eliminate trials with latencies > 10,000 ms");

The script does that. Trials with latencies greater than 10.000 ms are not included in the calculation of expressions.d.