Millisecond Forums

You'll find this covered in the comments of the Multifactor IAT script:

"D-scores obtained with this script:
Positive scores indicate a preference for the lefthand category
Negative scores indicate a preference for the righthand category

Example:
expressions.ABd (A is on the left; B is on the right) => A (=Christianity) vs. B (=Islam)
positive D-score indicates a preference for Christianity over Islam; negative D-score indicates a preference for Islam over Christianity.

in this script: categories
A: Babies
B: Puppies
C: Kittens
D: Pandas"

I.e., in your case this means:
positive ABd = preference for "culture 1" over "culture 2" / negative ABd = preference for "culture 2" over "culture 1"
positive ACd = preference for "culture 1" over "culture 3" / negative ACd = preference for "culture 3" over "culture 1"
positive ADd = preference for "culture 1" over "culture 4" / negative ADd = preference for "culture 4" over "culture 1"
positive BCd = preference for "culture 2" over "culture 3" / negative BCd = preference for "culture 3" over "culture 2"
positive BDd = preference for "culture 2" over "culture 4" / negative BDd = preference for "culture 4" over "culture 2"
positive CDd = preference for "culture 3" over "culture 4" / negative CDd = preference for "culture 4" over "culture 3"

Sorry if I am being daft, but I don't understand where do the attribute categories sit within these d-scores? 

Sorry, I'm afraid I don't understand your question. D-scores are based on response latencies, and targets are paired with attributes to the same response keys.

Indeed. d is calculated as one pair target and attribute response latencies minus the second pairing, divide by inclusive standard deviation, right?

For example, good babies versus bad adults - bad babies versus good adults / SD.

Where as you say:

expressions.ABd (A is on the left; B is on the right) => A (=Christianity) vs. B (=Islam)
positive D-score indicates a preference for Christianity over Islam; negative D-score indicates a preference for Islam over Christianity.

Don’t you need to include the pairings in your example to be able to calculate preference?

Again apologies if I’m missing something, I’m replying on my phone which is less than ideal!

> Indeed. d is calculated as one pair target and attribute response latencies minus the second pairing, divide by inclusive standard deviation, right?

Yes, that's the basic reasoning underlying D-scores. In essence, they're similar to typical effect size measures such as Cohen's d. As you rightly say, the fundamental definition is

[meanRT(incompatible pairing) - meanRT(compatible pairing)] / standard deviation

> Don’t you need to include the pairings in your example to be able to calculate preference?

But they are implicitly included? The Multifactor IAT, which is an extension of the Brief IAT, works like this (all of this is covered in the script's comments / manual):

For the test, participants are asked to sort categories into  paired/combined categories (e.g. "Infants OR Good" on the left vs. "Anything else" on the right). The basic task is to press a left key (E) if an item (e.g. "joyful" or picture of a human infant) belongs to the category presented on the left (e.g. "Infants OR Good") and to press the right key (I)
if the word (e.g. "tragic" or picture of a puppy) does not belong to the category on the left. Pairings are reversed for a second test (e.g. "Puppies OR Good" on the left vs. "Anything else" on the right). Each mammal offspring is tested against each other.

So, what you end up with is a set of paired comparisons:

Block AB: "Infants OR Good" vs. "anything else" (aka Puppies OR bad words)
Block BA: "Puppies OR Good" vs. "anything else" (aka Infants OR bad words)

Block AC: "Infants OR Good" vs. "anything else" (aka Kittens OR bad words)
Block CA: "Kittens OR Good" vs. "anything else" (aka Infants OR bad words)

Block AD: "Infants OR Good" vs. "anything else" (aka Panda cubs OR bad words)
Block DA: "Panda cubs OR Good" vs. "anything else" (aka Infants OR bad words)

Block BC: "Puppies OR Good" vs. "anything else" (aka Kittens OR bad words)
Block CB: "Non-binary gender presentation OR Good" vs. "anything else" (aka Puppies OR bad words)

Block BD: "Puppies OR Good" vs. "anything else" (aka Panda cubs OR bad words)
Block DB: "Panda cubs OR Good" vs. "anything else" (aka Puppies OR bad words)

Block CD: "Kittens OR Good" vs. "anything else" (aka Panda cubs OR bad words)
Block DC: "Panda cubs OR Good" vs. "anything else" (aka Kittens OR bad words)

And the D-scores are calculated as usual:

/ ABd = (BAm - ABm) / expressions.ABBAsd
/ ACd = (CAm - ACm) / expressions.ACCAsd
/ ADd = (DAm - ADm) / expressions.ADDAsd
/ BCd = (CBm - BCm) / expressions.BCCBsd
/ BDd = (DBm - BDm) / expressions.BDDBsd
/ CDd = (DCm - CDm) / expressions.CDDCsd