Difference between revisions of "Data analysis - StarJack"

Revision as of 23:26, 9 June 2011

The game begins with unparameterised SJ_START (1) and ends with SJ_END (19) followed by SJ_ENDSCREEN_CONTINUE (7). The game also is considered to have ended if the begin code for any other mini-game occurs. The game comprises multiple iterations of three phases: the maze task (which can be used as a measure of executive function), the Embedded Figures Test (where reaction times measure perceptual disembedding), and the "Sally-Anne" test of theory-of-mind.

Each level of the maze task begins with parameterised SJ_START (2) and ends with parameterised SJ_END (19). Each level of the maze task is interrupted by invocations of the Embedded Figures Test, and punctuated by an invocation of the Sally-Anne test. These latter two tests also can be invoked directly from the menu, without going through the maze test.

The Embedded Figures Test begins with SJ_EFT_START (49) and ends with SJ_EFT_END_FAIL (44) or SJ_EFT_END_SUCCESS (50).

The Sally-Anne test begins with SJ_SAT_START (52) and ends with SJ_SAT_END (53).

Following is a list of all the event codes, then a summary of the relevant behavioural and electrophysiological derived measures to extract:

After deletion of codes 255 and greater, and alignment of the behavioural and EEG event sequences just as was done for Maritime Defender, processing should continue by correcting (both in the behavioural log file and in the EEG log file) the numeric event codes that are associated with the following string event descriptors in the behavioural log file (see notes in the table above):
SJ_SAT_FirstOrder_NoTheft_Incorrect
SJ_SAT_FirstOrder_Scan_Correct
SJ_SAT_SecondOrder_NoTheft_Incorrect
SJ_SAT_SecondOrder_Scan_Correct
SJ_SAT_SecondOrder_Question_Presented
SJ_SAT_VideoStart_Cruiser_View_Theft

SJ_EFT_INPUT_Match and SJ_EFT_INPUT_NoMatch events should be re-coded as context-free events based on the success flags (SJ_EFT_HIT, SJ_EFT_REJECT, SJ_EFT_FALSE_ALARM, SJ_EFT_MISS) that follow each of them. The success flags themselves then can be deleted, as they do not mark any actual events distinct from the SJ_EFT_INPUT_Match SJ_EFT_INPUT_NoMatch that they follow.

Similarly, SJ_SAT_INPUT_Response_{Top, Bottom, Left, Right} events should be re-coded as context-free events based on the success flags (SJ_SAT_FirstOrder_{*} and SJ_SAT_SecondOrder_{*}). In the re-coded events, both the four possible locations of the response (Top, Bottom, Left, Right) and the four possible correctness conditions of the response (Correct, IncorrectTrueLocation, IncorrectFakeLocation, IncorrectOtherLocation) should be preserved, thus there will be 4x4 = 16 possible codes. The success flags themselves then can be deleted, as they do not mark any actual events distinct from the SJ_SAT_INPUT_Response_{*} events that they follow.

Behavioural measures related to executive function

As simple behavioural measures of executive function, the software can compute - for each level separately and for all levels cumulatively - the total number of moves, the total time elapsed from start to finish (excluding time spent in EFT and SAT subtests), and the mean and standard deviation of the time between successive moves. Other measures are possible; if you think of one, try it!

Behavioural measures related to perceptual disembedding

d' ("d prime") should be computed from tallies of EFT hits, misses, false alarms and correct rejections. If the tally for any one of these four categories of responses is zero, that tally can be changed to 1 so as to estimate d'.

Mean and standard deviation of EFT reaction time, for correct responses only, should be computed. Also list out the single-trial reaction times.

Behavioural measures related to theory-of-mind

Single-trial data should be extracted and listed out for first-order and second-order questions in all trials. Three types of trials exist: the first-order control condition in which the pirate scans but does not steal the cargo (probability 1/3, and note that in this case it doesn't matter whether or not the admiral observes the pirate; the observed and unobserved cases can be collapsed into a single bin), the first-order theory-of-mind conditions in which the pirate steals the cargo but is not observed by the admiral (probability 1/3), and the second-order theory-of-mind condition in which the pirate steals the cargo but the admiral observes this theft and tells the captain (probability 1/3). The single-trial data should be listed first for the no-theft condition, then separately for the unobserved theft condition, then separately for the observed theft condition. The data for each trial comprise three timings and two correctness values: the time taken for the whole trial (SJ_SAT_END minus SJ_SAT_START), the reaction time for response to the first-order question (SJ_SAT_INPUT_Response_{*} minus SJ_SAT_FirstOrder_Question_Presented) and the correctness condition (Correct, IncorrectTrueLocation, IncorrectFakeLocation, IncorrectOtherLocation) of that response, and the reaction time for response to the second-order question (SJ_SAT_INPUT_Response_{*} minus SJ_SAT_SecondOrder_Question_Presented) and the correctness condition of that response.

Summaries of correctness conditions for each of the two question types (first order, second order) for each of the three theft conditions (no theft, unobserved theft, observed theft) should be given both as tallies and as percentages.

For correct responses only, means and standard deviations (where computable) should be given for the following reaction-time differences:

First-order question:
Unobserved Theft minus No Theft
Unobserved Theft minus Observed Theft
Unobserved Theft minus (No Theft or Observed Theft, pooled)

Second-order question:
Observed Theft minus Unobserved Theft
Observed Theft minus No Theft
Observed Theft minus (Unobserved Theft or No Theft, pooled)

Independent Components Analysis

ICA can be applied separately to epochs from the Embedded Figures Test, epochs from the Sally-Anne test, and perhaps continuous EEG for the maze task.

Remember not to include the skin-conductance channel (the greatest-numbered channel) in the ICA!

Physiological measures related to perceptual disembedding

Standard ERP and ERSP analyses can be constructed round the SJ_EFT_Stimulus_Presented event and corresponding SJ_EFT_INPUT_{*} response events.

Physiological measures related to theory-of-mind

Some possible fMRI-based contrasts:

Gaze perception (blink events or gaze aversion intervals versus direct gaze intervals)

Facial expression perception (scowl intervals versus neutral intervals)

Scene perception (egocentric view of person's face, versus allocentric view of person's hands on console, versus external view of spaceships)

Perception of body movements (hands) or gestures (nods)

Also note that, because these stimuli are cinematic, useful information can be gleaned from applying Uri Hasson's approach of correlating individual subjects' voxelwise activation time series against the group mean. This model-free procedure can identidy consistent activations, localised temporally and anatomically, across subjects. The timings of such consistent activities then can be correlated with the video timecodes (see link in the table above).

It may be useful to contrast fMRI time series during the video stimuli analogous to the behavioural (reaction-time) contrasts detailed above - that is, Unobserved Theft minus No Theft, Unobserved Theft minus Observed Theft, and so forth. (For a list of specific contrasts, see the behavioural section above.)

For EEG analyses, given the temporally extended nature of these stimuli spectral and other quantitative-EEG methods may be more appropriate than ERP's.

Physiological measures related to executive function

Here again, quantitative-EEG methods may be most appropriate. In addition, specific time-locked events may generate specific ERP's of interest, for example error-related negativity on SJ_INPUT_Move_{*} events that immediately precede SJ_Caught_TeleportToStart events.