People are more likely to form memories for high than low reward items, and are faster to recognize high reward items as old.
There are no age group differences in the influence of rewards on memory accuracy or memory RT
There is an age x sex x reward size interaction; this is driven by age differences in the influence of reward on memory in females but not males
There is an age x reward size x post-jitter time interaction: when people are given a longer time to boost memory after an encoding trial, adolescents have more selective memories for high reward items than adults
I first checked whether overall d prime differed by age group and whether there were any participants who needed to be excluded due to particularly low memory performance. Unfortunately, 8 participants have d prime score below zero. I’m plotting d prime on the left with everyone included and on the right restricted to those with d prime above zero. Given it was an intentional memory task, I’m surprised d prime is so low. For the rest of the analyses, I’ll report results when including the full sample, as well as analyses restricted to those who show d prime scores > 0.
memory hit = either familiar or remember response to old item
false alarm = either familiar or remember response to new item
full sample: b = −0.19, SE = 0.06, t(47) = −3.18, p = .003
d prime > 0 subgroup: b = −0.11, SE = 0.05, t(38) = −2.04, p = .05
I tested whether people were more likely to remember high than low reward items and whether this differed by age group (age x reward size interaction). On average, people had better memory for high than low reward items:
full sample: b = 0.12, SE = 0.03, z = 4.13, p < .001
d prime > 0 group: b = 0.13, SE = 0.03, z = 4.18, p < .001
Unfortunately (for us!) this did not differ significantly by age group
full sample: b = −0.05, SE = 0.03, z = −1.62, p = .106
d prime > 0 subgroup: b = −0.04, SE = 0.03, z = −1.32, p = .187
People have higher memory d prime for images associated with high than low rewards (full sample: b = 0.08, SE = 0.02, t(48) = 4.28, p < .001; dprime > 0 subgroup: b = 0.08, SE = 0.02, t(40) = 4.12, p < .001) but no age group interaction (full sample: b = −0.02, SE = 0.02, t(48) = −1.41, p = .166; dprime > 0 subgroup: b = −0.02, SE = 0.02, t(40) = −0.99, p = .328).
Given overall memory differed by sex, I fit a model that included sex as a covariate and interaction term (age x reward size x sex interaction). There was a 3 way interaction between age group x reward size x sex:
full sample: b = 0.05, SE = 0.03, z = 1.73, p = .083
dprime > 0 subgroup: b = 0.07, SE = 0.03, z = 2.18, p = .029
This was driven by larger age differences in memory for high than low rewards in females (reward size x age group interaction: b = −0.10, SE = 0.04, z = −2.35, p = .019) than males (reward size x age group interaction: b = 0.003, SE = 0.04, z = 0.08, p = .936).
I found a 3 way interaction between age x sex and reward size in participants who have d prime scores above zero:
full sample: b = 0.03, SE = 0.02, t(46) = 1.54, p = .131
d prime > 0 subgroup: b = 0.04, SE = 0.02, t(38) = 1.97, p = .056
This is driven by larger age differences the the influence of rewards for females (b = −0.05, SE = 0.03, t(38) = −2.02, p = .051) than males (b = 0.03, SE = 0.03, t(38) = 0.86, p = .394).
People endorsed high reward items as old faster than low reward items:
But this didn’t differ by age-group:
full sample: b = 0.002, SE = 0.005, t(3407) = 0.35, p = .723
d prime scores above zero: b = 0.003, SE = 0.005, t(3050) = 0.61, p = .542)
Furthermore, there are no age x sex x reward size interactions (ps > 0.84). Only correctly recognized items are included here.
I looked at whether having a longer post item jitter exacerbated differences in memory for high and low reward items. The idea being that a longer time between seeing an item and the next might allow individuals more time to upregulate memory via active rehearsal. I only included participants with d prime scores above zero here.
There was a reward size x age group x post jitter time interaction (b = −0.06, SE = 0.03, z = −1.98, p = .047).
Longer post item jitter times led to a greater differences in memory between high and low reward items in adolescents than adults
long: age x reward size: b = −0.11, SE = 0.05, z = −2.33, p = .020
short: age x reward size: b = 0.05, SE = 0.06, z = 0.75, p = .452
This suggests that having time after we encode something can augment age differences in the influence of reward. This effect did not interact with sex.
I looked at whether receiving a reward or punishment during the recognition phase influenced memory accuracy on the next trial and whether this differed by age/sex. The answer is no, punishments/rewards during recognition do not impact accuracy on the subsequent recognition trial. This suggests that the effects of reward on memory are most “potent” during encoding and can’t help you after you’re formed a memory. Sex doesn’t modulate the effects. Here, I’ve plotted memory hits (old items only), but the results are similar when I look at memory accuracy (both old and new items).
I also looked at whether time on task modulated anything and found a 3 way interaction between preceding reward X time on task X group (ps < .009). The pattern is similar when I model time on task continuously. Higher rewards boost memory on the NEXT trial more in adolescents than adults in the first half of the recognition phase; For adolescents, this pattern flips in the second half.
I looked at whether the effects of reward during encoding were larger in the first than second half because of waning attention during encoding. I split the encoding data into two halves and fit an age group x trial type x task half interaction. There was an age x reward size x task half interaction in the full sample, but not after excluding the subgroup of participants whose d prime scores fell below zero:
full sample: b = 0.06, SE = 0.03, z = 2.19, p = .029
dprime scores > 0: b = 0.04, SE = 0.03, z = 1.17, p = .243
This was driven by a larger age differences in the influence of reward in the first (b = −0.11, SE = 0.04, z = −2.70, p = .007) than second half of the task (b = 0.02, SE = 0.04, z = 0.38, p = .703).
I was analyzing the data using the text files located on mindhive; there were two participants (adult participants #1 and #5) whose recognition data had items repeated more than once; I did not exclude these participants but just took the first instance that a particular image showed up during recognition and removed the repeat from the data.
There was also an adolescent participant (#24) who did not have a full recognition data. They were missing a few trials in run 2.