CNP BART

From Pheno Wiki
Revision as of 10:43, 20 May 2011 by AndyD (Talk | contribs)

Jump to: navigation, search

go back to HTAC


Basic Task Description

The Balloon Analogue Risk Task (BART) is a computerized behavioral measure of risky decision-making, developed by Carl Lejuez (Lejuez et al., 2002). Balloons are presented on a computer screen, one balloon per trial, and the examinee can “pump” the balloons up by pressing a response key, virtually inflating the balloons (i.e., they increase in size). Each time a balloon is pumped up, a set amount of money or points is earned on that trial (e.g., 5 cents per pump). However, after a certain number of pumps (determined probabilistically), the balloons explode and no money or points are earned on that trial. Thus, the examinee must decide when to “cash out” of a given trial to put the trial money/points into a cumulative bank, by pressing a “cash out” response key. If a balloon explodes prior to cashing out, no money is earned on that particular trial. The objective is for the examinee to earn as much money or points as possible across the task. Versions of the BART vary with respect to the number of trials/balloons used, as well as the probability of explosions [e.g., some tasks have used balloons with a single probability of explosion (e.g., Lejuez et al., 2003), while others have used different-colored balloons with different probabilities of explosion (e.g., Lejuez et al., 2002; Dean et al., 2011)]. The primary dependent variable of the task is the mean or total number of pumps on trials in which the balloon did not explode; these have been termed adjusted pumps. Adjusted pumps are preferred to absolute pumps because explosions artificially restrict the range of pumping behavior (for evidence of the bias associated with absolute pumps see Pleskac et al. 2008).

Task Procedure

For general testing procedure, please refer to LA2K General Testing Procedure [here?].

Our version of the BART task was programmed in E-Prime 2.0. It consists of 40 total trials, with balloons that are colored red or blue (20 of each color). Red balloons are relatively “high risk”; they have a 1/32 probability of exploding on the first pump, and an incrementally ascending probability of explosion with each successive pump (1/31 on the second pump, 1/30 on the third pump, etc.). Blue balloons are “low risk”, with a 1/128 probability of exploding on the first pump (1/127 on the second pump, etc.). The order of balloon color across trials was randomized. Examinees are given 5 points for each adjusted pump. In our version of the task, participants are not paid for their performance.

At the beginning of the task, instructions are presented on the screen in yellow, size 16 Arial font, with a black screen background. In order to reduce learning affects on pumping behavior, the examinees are informed at the outset that two different balloons are used with different explosion tendencies (see Wallsten et al., 2005 for consideration of this issue). The instructions are as follows:

  • In the following game, you pump up a balloon.
  • Your goal is to earn as many points as possible.
  • You can get 5 points for each pump!
  • However, when the balloon explodes, you get no points.
  • You can save your points by choosing to stop pumping before the balloon explodes. There are two different balloons and they are different in how much they can be inflated before they explode.
  • There are a set number of trials in this experiment, so try to maximize your points on each trial.

These instructions are reiterated on following screens, in which the buttons used for pumping and cashing out are indicated. If the subject is right-handed (this information is entered into the program during start-up), the pumping key is the left key on a 3-button keypad, while the cash-out key is the right key. In contrast, if the examinee is left handed, this button mapping is reversed (i.e., pumping with the right key, and cashing-out with the left key). The placement of the keypad is also oriented on the right or left side of the examinee in accordance with handedness.

After the examinee expresses understanding of the task, the experimental session begins. A blank screen with a fixation cross is briefly shown (500 milliseconds), and then the first balloon is presented (red or blue). Text is placed below the balloon image to indicate which key press is used for pumping and which is used for cashing in (i.e., serving as a reminder). Balloons are presented centrally on an all white screen background. The first balloon image is 152 X 152 pixels in size. After one pump, another balloon image is presented instantaneously with a size of 154 X 154 pixels. The balloon size increases in height and width by 2 pixels with each successive pump, giving the appearance of being inflated. This continues until the examinee cashes out or the balloon explodes, whichever occurs first. Upon explosion, an image of a ruptured balloon is presented of comparable size to the last balloon image. After both cash outs and explosions, a rectangular yellow text box is presented on the screen for 1500 milliseconds which shows the examinee’s current Total Points in black text (Arial size 16 font). In the case of an explosion, this total will be identical to the previous trial’s total because no points have been earned*. Note that accumulated points are only shown at the end of a given trial, and are not shown as the examinee is pumping. After the previous trail is concluded, a blank screen with a central fixation cross is presented for 500 milliseconds, followed by presentation of the next balloon/trial. This process continues until the 40th balloon is completed. Afterward, a black screen is presented which thanks the participant for his/her participation, and shows a yellow text box with the Grand Total Points earned (Arial size 16 font, yellow text). The task is self-paced and duration depends on the examinee’s behavior, but initial results from our data indicate that the task took an average of XXXXXXXXXXX minutes across participants.

  • Because of a programming error, in some cases examinees will receive points on balloons which exploded. It is thus imperative that points not be used as a dependent variable. Rather, adjusted pumps should be used.

Task Structure Detail

This is what we had worked on before, but could use updating. We'd like to capture a schema that can handle each of the tasks in the CNP, so please think general when editing -fws

  • Task Structure (please given an overview of the task procedures here [i.e., overall design, block, trial, and within-trial event structure and timing])
    • The BART has a series of instructional screens at the outset, followed by the experimental session comprised of 40 balloon trials.
      • Six instructional screens. Screens are advanced with a right mouse click from the examiner.
        • 1. Basic task description.
        • 2. Placement of hands and keys used for pumping and stopping (mapped based on handedness).
        • 3. Pumping description and appearance of balloon.
        • 4. Explosion description and appearance of exploded balloon.
        • 5. Reiteration of key used to stop pumping/cash in (mapped based on handedness).
        • 6. Questions screen prior to beginning task.
      • Experimental trials (40 trials). All experimental trials are preceded by a fixation cross screen.
        • Fixation cross (500 ms) screen with blank white background.
        • Balloon trial 1. Red or blue balloon presented (randomized).
          • Within-trial structure
            • Pumping increases balloon size with successively enlarged images (self-paced).
            • Participant cashes out or the balloon explodes.
            • Total points earned screen presented for 1500 ms.
        • Balloon trial 2 begins after a fixation cross screen, and this is sequence is repeated for 40 trials.
      • End. Thank you screen presented. Grand Total Points also displayed.
    • Timing
      • Instruction screens are static until advanced by examiner with a right mouse click.
      • Fixation cross screens (500 ms).
      • Balloon images are static until participant presses the pump key. Pressing the pump key instantaneously presents the following balloon image (either a slightly larger balloon or an exploded balloon).
      • Images of an exploded balloon are presented for 1500 ms, with Total Points text underneath.
      • When the participant cashes out, just the Total Points text is presented for 1500 ms.
      • The ending thank you screen is static until advanced by the examiner with right mouse click.
  • Stimulus Characteristics
    • sensory modality: Visual. Balloons are either red or blue. Balloon images start with a size of 154 X 154 pixels and are increased in height and width by 2 pixels with each pump. The fixation cross is black text in size 18 bold font in Comic Sans MS.
    • functional modality: visuoperceptual and linguistic (understanding of text).
    • presentation modality: computer display, no audio, directions are assisted by examiner.
  • Response Characteristics
    • responses required: left or right key press for pumping or cashing in. Button mapping is based on handedness.
      • effector modality: Manual button press.
      • functional modality: Manual button press.
    • response options (e.g., yes/no, go/no-go, forced choice, multiple choice [specify n of options], free response): yes/no, pump or cashout.
    • response collection (e.g., examiner notes, keyboard, keypad, mouse, voice key, button press): Button press and recording of responses in Eprime 2.0.

Task Schematic

Schematic of the Balloon Analogue Risk Task

Total Points: XX

    Total Points: XX


Task Parameters Table

TaskParamTable.png

Stimuli

Stimuli consist of red or blue balloons which begin with a size of 154 X 154 pixels and are increased in height and width by 2 pixels with each pump. When a balloon explodes, an image comparable in size and color is displayed as a burst balloon for 1500 ms.

A white screen with a black fixation cross (size 18 bold font in Comic Sans MS) is presented for 500 ms before every balloon trial.

Dependent Variables

The primary dependent variable is the mean number of pumps on trials in which the balloon did not explode; these have been termed adjusted pumps. Adjusted pumps are preferred to absolute pumps because explosions artificially restrict the range of pumping behavior (for evidence of the bias associated with absolute pumps see Pleskac et al. 2008).

Because the red and blue balloons differ considerably in their probability of explosion, it may also be useful to analyze the mean adjusted pumps on the red and blue balloons separately. In addition, because participants will likely adapt to the task over time, adjusted pumps for each balloon color on each quartile of trials (1-10, 11-20, 21-30, 31-40) can be analyzed to determine how performance changes over time. Note here that mean adjusted pumps for each quartile likely vary in terms of the number of trials averaged (due to explosions and the fact that balloon color is randomized across the task). Other variables of interest may include the number of adjusted pumps on red and blue balloons which immediately follow an explosion, and the number of explosions for each balloon color.


Other dependent variables that may be of interest include

Additional summary measures can be used to screen outliers (see below).

Table of all available variables.

SST Variables Table.png

Cleaning Rules

If any of the derived variables (those listed in variables Table above) are missing, for either one or both testing blocks, that participants should be flagged for exclusion.

There are several decisions to make when estimating SSRT from more than one block of Stop-signal task performance data, including whether to average across all available sessions or to use the last session run (based on the assumption that participants are closest to their 50% inhibition point at the end of a session); whether to use all trials of each session or the last half only (again based on the assumption that participants stabilize inhibitory performance near the end of a session); and whether to use data from all participants that completed the task regardless of performance, or to use either conservative or lenient criteria to exclude outliers, so as to avoid violating assumptions underlying the race-model of stopping. These questions have been systematically assessed in an independent dataset, which was randomly split into halves in order to evaluate reliability and repeatability of SSRT estimates derived following multiple approaches to data cleaning (Congdon et al., in preparation). Measures of reliability, including intraclass correlation coeffcients (ICC) and within-subject variability, and the resulting sample size, were used as indicators to evaluate the different strategies to data cleaning.

Our results suggest that an approach that uses the average of all available Stop-signal blocks, all trials of each block, and excludes outliers based on predetermined lenient criteria (defined below) yields reliable SSRT estimates and low within-subject variability, while not excluding too many participants from the total dataset. Specifically, this approach resulted in an ICC value of 0.79 and a within-subject variability estimate of 25.42 ms, while only excluding 7 (out of 184) participants. Critically, this approach also retains a broad distribution of SSRT values.

Based on these analyses, the following cleaning rules are suggested:

  • Use all trials from each testing block
  • Average across all available testing blocks for final summary scores
  • Exclude outliers that meet the following lenient criteria:
    • Percent inhibition on Stop trials less than 25% or greater than 75%
    • Percent correct responding on Go trials less than 60%
    • Percent incorrect Go trials greater than 10%
    • SSRT estimate that is negative or less than 50 ms

Code/Algorithms

Scoring of behavioral data proceeded as follows.

The mean, median and standard deviation of reaction time on Go trials were calculated only for Go trials in which participants correctly responded. Stop successful trials included only Stop trials on which participants successfully inhibited a response, and Stop unsuccessful trials included only Stop trials on which participants responded. Average stop-signal delay (SSD) was calculated from SSD values across both ladders. SSRT was estimated using the quantile method, which does not require an assumption of 50% inhibition (Band et al., 2003). Briefly, to calculate SSRT following the quantile method, all correct RTs from Go trials were arranged in ascending order. The proportion of failed inhibition, which is the proportion of Stop trials in which the participants responded, was calculated across both ladders. The quantileRT was determined by finding the RT corresponding to the proportion of failed inhibition. The average stop-signal delay (across both ladders) was then subtracted from the quantileRT in order to calculate SSRT (Band et al., 2003).

Scores are calculated from each block separately and then averaged to provide summary scores for the overall session. It is recommended that one use the final measures based on the overall session, as they provide more stable estimates of SSRT (Band et al., 2003; Congdon et al., in preparation).


Make 2 Filters: Procedure[Block] and Running[Block] and Procedure[Trial]

First filter to only include trials where Procedure[Block] = “StopProc” (these are the real trials)

Analyze the remaining trials in 2 different sets: Those with Running[Block] = “Block1” and those with Running[Block] = “Block2”

For each of those two sets (Block1 and Block2) do the following steps:

{

  • Block1_Direction_Errors = Number of trials where (Procedure[Trial] = "StGTrial" AND Go.RT> 0 AND Go.ACC= 0) OR (Procedure[Trial] = "StGTrial" AND Go.RT = 0 AND Blank.Resp != CorrectAnswer)
  • Block1_Percent_Go_Response = [Number of trials where Procedure[Trial] = "StGTrial" AND (Go.ACC=1 OR Blank.ACC=1)] / Number of trials where Procedure[Trial] = "StGTrial"

For the next calculation, you need the following values:

  • Go.RT for all trials where (Procedure[Trial] = "StGTrial") AND (GoAcc = 1 AND Go.RT > 0)
  • Blank.RT + 1000 for all trials where (Procedure[Trial] = "StGTrial") AND (GoACC = 0 AND Go.RT = 0 AND Blank.ACC = 1)
  • Block1_Mean_RT = Mean of all values that you just got from Go.RT and Blank.RT + 1000
  • Block1_Median_RT = Median of all values that you just got from Go.RT and Blank.RT + 1000
  • Block1_StDev_RT = Standard deviation of all values that you just got from Go.RT and Blank.RT + 1000

For the next step, you need to get have these numbers:

  • GoDur - 50 for all trials where Procedure[Trial] = "StITrial" AND (Go1s.RT=0 and Inhs.RT=0 and Go2S.RT=0 and Blanks.RT=0)
  • GoDur + 50 for all trials where Procedure[Trial] = "StITrial" AND (Go1s.RT!=0 or Inhs.RT!=0 or Go2S.RT!=0 or Blanks.RT!=0)
  • GoDur2 - 50 for all trials where Procedure[Trial] = "StITrial2" AND (Go1s2.RT=0 and Inhs2.RT=0 and Go2s2.RT=0 and Blanks.RT=0)
  • GoDur2 + 50 for all trials where Procedure[Trial] = "StITrial2" AND (Go1s2.RT!=0 or Inhs2.RT!=0 or Go2s2.RT!=0 or Blanks.RT!=0)
  • Block1_Ladder1Mean = Mean of all GoDur values from the previous statements (there should be 16 total)
  • Block1_Ladder2Mean = Mean of all GoDur2 values from the previous statements (there should be 16 total)
  • Block1_SSD50 = Mean of ALL the values you just got from the previous statements (there should be 32 total values that you're taking the mean of)


  • Block1_PctInhib_Ladder1 = [Number of Trials where Procedure[Trial] = "StITrial" AND (Go1s.RT=0 and Inhs.RT=0 and Go2s.RT=0 and Blanks.RT=0)] / (Number of Trials where Procedure[Trial] = "StITrial")
  • Block1_PctInhib_Ladder2 = [Number of Trials where Procedure[Trial] = "StITrial2" AND (Go1s2.RT=0 and Inhs2.RT=0 and Go2s2.RT=0 and Blanks.RT=0)] / (Number of Trials where Procedure[Trial] = "StITrial2)
  • Block1_Percent_Inhib = Mean of Block1_PctInhib_Ladder1 and Block1_PctInhib_Ladder2
  • Block1_Quantile_Value = 1 - Block1_Percent_Inhib
  • Block1_SSRT = Block1_Median_RT - Block1_SSD50
  • CorrectGoRT on StopProc trials: As a reminder, these are Go.RT for all trials where (Procedure[Trial] = "StGTrial") AND (GoAcc = 1 AND Go.RT > 0), and Blank.RT + 1000 for all trials where (Procedure[Trial] = "StGTrial") AND (GoACC = 1 AND Go.RT = 0). Use these CorrectGoRT on StopProc trials to calculate Block1_Quant_RT:
  • Block1_Quant_RT = Quantile calculation of all the values from Go.RT and Blank.RT + 1000 (the values used in the calculations where you calculated Block1_Mean_RT, etc), taken using Quantile_Value. So if Quantile_Value = .85, you'd take the .85-quantile of the distribution of values from Go.RT and (Blank.RT + 1000)
  • Block1_SSRT_Quant = Block1_Quant_RT - Block1_SSD50

}

Then, do all those same calculations that were in the braces, but using the data points that were from Running[Block] = “Block2” (and save the variables as Block2_whatever, rather than Block1)

Then average the above values together to get a measure of overall task performance:

All the values that we need outputted are in bold. See also Variable List.


The following text instructions can be adapted for any program. These instructions are currently implemented in Stone’s scoring scripts for the scoring of LA2K variables. These steps have been implemented in separate Matlab scripts by Eliza and have been used to verify Stone’s scoring scripts. These steps have also been used to manually score the data and verify Stone’s scoring scripts multiple times (finalized April 2010).

History of Checking Scoring:

  • October 2009: Eliza confirmed Stone’s scripts by manually scoring data (completed 10/19/2009)
  • January 2010: Another group (Nicole McLaughlin at Butler Hospital) that was sent the scoring scripts identified a scoring error.
    • Eliza and Stone worked together to resolve the problem. Stone fixed that error in the script on 1/29/2010.
    • Eliza and Stone clarified an issue about the scoring of Go trial (direction) errors and matched scores between UCLA and Butler (March 2010).
    • Final questions about differences in quantile calculation were resolved in April. Eliza and Stone calculated all scores on the same data and met to confirm scoring scripts. Everything agreed and finalized 4/21/10.

Note that one thing that may differ in the scoring script between programs is quantile calculation. For example, the quantile function in Matlab differs slightly from the percentile function in Excel. Stone’s scoring method may differ by a tiny fraction from the Matlab quantile function; however, this difference is tiny, and as long as the same scoring script is applied to all subjects (as it is in LA2K), then it does not matter. All agreed April 2010 on this issue.

Data Distributions

1 Results.png

2 Results.png

3 Results.png

4 Results.png

5 Results.png

References

Band, G. P., van der Molen, M. W., and Logan, G. D. (2003). Horse-race model simulations of the stop-signal procedure. Acta Psychol, 112, 105-142.

Boucher, L., Palmeri, T. J., Logan, G. D., and Schall, J. D. (2007). Inhibitory control in mind and brain: An interactive race model of countermanding saccades. Psychol Rev, 114 (2), 376-97.

Logan, G. D., and Cowan, W. B. (1984). On the ability to inhibit through and action: a theory of an act of control. Psychol Rev, 91, 295-327.

Logan, G. D. (1994). On the ability to inhibit thought and action: A users’ guide to the stop signal paradigm. In: Dagenbach, D., Carr, T. H. (Eds.), Inhibitory Processes in Attention, Memory and Language. Academic Press, San Diego, pp. 189-239.