CRP Tutorial

From Computational Memory Lab
Jump to: navigation, search

Single-trial example

We will label presented items (for our lab, usually words) with numbers 1-8, in sequence, so our hypothetical participant sees this string of items, one at a time:

1 2 3 4 5 6 7 8

The participant is told to recall the presented items in any order, and then recalls items corresponding to the following numbers, in this order:

8 7 1 2 3 5 6 4

Let's first tally up the transitions that were actually made:

  1. 8 to 7: lag of -1
  2. 7 to 1: lag of -6
  3. 1 to 2: lag of +1
  4. 2 to 3: lag of +1
  5. 3 to 5: lag of +2
  6. 5 to 6: lag of +1
  7. 6 to 4: lag of -2

This gives us an actual transition tally that looks like this:

Lag -7 -6 -5 -4 -3 -2 -1 +1 +2 +3 +4 +5 +6 +7
Actual count 0 1 0 0 0 1 1 3 1 0 0 0 0 0

Note that the maximum absolute lag possible with a list of 8 items is 8-1=7 (i.e., a transition from first to last or vice versa).

Because we want a conditional response probability, we also have to tally up the transitions that could have been made at each output. The list is long, so please click "Expand" below to show:

Possible transitions

  1. After the first output (item 8), the participant could have recalled:
    1. Item 1: lag of -7
    2. Item 2: lag of -6
    3. Item 3: lag of -5
    4. Item 4: lag of -4
    5. Item 5: lag of -3
    6. Item 6: lag of -2
    7. Item 7: lag of -1
  2. After the second output (item 7), the participant could have recalled:
    1. Item 1: lag of -6
    2. Item 2: lag of -5
    3. Item 3: lag of -4
    4. Item 4: lag of -3
    5. Item 5: lag of -2
    6. Item 6: lag of -1
    • Note that the participant can not make a valid transition of +1, since item 8 has already been recalled
  3. After the third output (item 1), the participant could have recalled:
    1. Item 2: lag of +1
    2. Item 3: lag of +2
    3. Item 4: lag of +3
    4. Item 5: lag of +4
    5. Item 6: lag of +5
    • Note that the participant can not make valid transitions of +6 or +7, since items 7 and 8 have already been recalled
  4. After the fourth output (item 2), the participant could have recalled:
    1. Item 3: lag of +1
    2. Item 4: lag of +2
    3. Item 5: lag of +3
    4. Item 6: lag of +4
    • Note that the participant can not make valid transitions of -1, +5 or +6, since items 1, 7 and 8 have already been recalled
  5. After the fifth output (item 3), the participant could have recalled:
    1. Item 4: lag of +1
    2. Item 5: lag of +2
    3. Item 6: lag of +3
    • Note that the participant can not make valid transitions of -1, -2, +4 or +5, since items 1, 2, 7 and 8 have already been recalled
  6. After the sixth output (item 5), the participant could have recalled:
    1. Item 4: lag of -1
    2. Item 6: lag of +1
    • Note that the participant can not make valid transitions of -4, -3, -2, +2 or +3, since items 1, 2, 3, 7 and 8 have already been recalled
  7. After the seventh output (item 6), the participant could only have validly recalled:
    1. Item 4: lag of -2
    • Note that this is the only valid transition left, since all other items have been recalled.
  8. After the eighth output (item 4), there are no more possible valid transitions
    • Note that repetitions are never included, since they do not count as valid transitions.
Resulting lag-CRP curve from single-trial example

Adding up all possible lags, we get a possible transition tally of:

Lag -7 -6 -5 -4 -3 -2 -1 +1 +2 +3 +4 +5 +6 +7
Possible count 1 2 2 2 2 3 3 4 3 3 2 1 0 0

We then divide the actual count by the possible count to get the conditional probability:

Lag -7 -6 -5 -4 -3 -2 -1 +1 +2 +3 +4 +5 +6 +7
Actual count 0 1 0 0 0 1 1 3 1 0 0 0 0 0
Possible count 1 2 2 2 2 3 3 4 3 3 2 1 0 0
Conditional Probability 0 0.50 0 0 0 0.33 0.33 0.75 0.33 0 0 0 0 0

This makes an extremely noisy lag-CRP curve (right), but this is expected with (a) such a short list, and especially (b) only a single trial.

Multi-trial example

This example is more advanced and assumes some programming knowledge, ideally in Matlab. (Ports of the examples to other languages are coming soon.)

The subject here is presented with X lists. In the below pres matrix, each row is a trial and each column is a serial position. The numbers are identifiers for the word that was presented (out of a 1638-item wordpool). For example, in the first trial, the subject was presented with word 1062, then 219, then 779, and so on. There are 16 columns, so the subject was presented with 16 items in each trial.

You may copy and paste the below code block into Matlab to create the pres variable.

pres = 
[1062  219  779  148  668 1458  694 1637  475 1433  912 1416  785 1411  701  969;
 1175  788  653 1031 1410  184 1134  993  152  313 1033 1591  652 1115 1184 1531;
  329  605  414 1614  852  950 1105   10 1144  459  856  455 1178  277 1415 1118;
  843 1173 1067   46 1435 1561 1202   30  270  493  815  102  233  796  500  753;
 1044  372  556 1600  685 1598  409  989 1326  100  757   94 1491 1398   56 1261;
  257 1454  315 1596 1400 1578  840 1069 1294 1254  572  758 1052  620 1230  380;
  546  811  252 1093  147  241  589 1351 1373 1446 1621  310  920  188   22  339;
 1179 1402 1468  581 1629  647  710  207 1313 1627  378  397  400  515 1579 1038;
 1107  410  206 1413  441  930 1439  975  242  109 1257 1193  535 1471  122  132;
 1427  555  949 1204  116  540  479  582 1386  888  258 1342 1198  574   87 1008];

After presentation of each list, the subject is given a period to recall the presented items in any order. In the below rec matrix, each row is a recall period following the corresponding presentations from the pres matrix. For example, in the recall period following the first trial's presented items, the subject recalled item 701, then 969, then 475, and so on. There are varying numbers of non-zero entries per row, indicating that the subject recalled a different number of items from the corresponding list in each trial, for example, 7 in the first trial, 5 in the second, and so on.

Zeros are placeholders indicating no recall. (The subject was given a fixed 75 second recall period to recall as many words as possible.)

You may copy and paste the below code block into Matlab to create the rec variable.

rec = 
[ 701  969  475  785 1411 1062  219    0    0    0    0    0    0    0    0    0;
 1531  313 1033  993 1134    0    0    0    0    0    0    0    0    0    0    0;
  605  414  950  852  277 1415  329  455 1591    0    0    0    0    0    0    0;
  500  753  843 1173   46  815    0    0    0    0    0    0    0    0    0    0;
 1261 1491  409  989 1044  372  556 1598  757    0    0    0    0    0    0    0;
  380  620 1052  315 1596  572 1294  758 1254    0    0    0    0    0    0    0;
  339   22  188 1373 1446  589    0    0    0    0    0    0    0    0    0    0;
 1038  515 1579  378    0    0    0    0    0    0    0    0    0    0    0    0;
  132 1471 1107 1193 1257  410  535  109 1439  930    0    0    0    0    0    0;
   87 1342 1008  574 1471  888  479  540   87  116    0    0    0    0    0    0];

The easiest way to construct a lag CRP is from a recalls matrix, which shows the serial position of recalled words. The simplest way to do that is to loop over trials/lists:

  1. nList = size(pres,1);       % Number of lists/trials; equal to number of rows in "pres"
  2. listLength = size(pres,2);  % Number of presented items per list;
  3.                             % equal to number of columns in "pres"
  4.  
  5. recalls = nan(nList,listLength);  % Preallocate with NaNs for speed
  6.  
  7. for i = 1:10 % Loop over trials
  8.     [~,recalls(i,:)] = ismember(rec(i,:),pres(i,:));  % Find each recalled item in the 
  9.                                                       % presented list and report its
  10.                                                       % (serial) position
  11. end
  12.  
  13. recalls  % Print the "recalls" matrix to the command window

Which should produce the following output:

recalls =

   15   16    9   13   14    1    2    0    0    0    0    0    0    0    0    0
   16   10   11    8    7    0    0    0    0    0    0    0    0    0    0    0
    2    3    6    5   14   15    1   12    0    0    0    0    0    0    0    0
   15   16    1    2    4   11    0    0    0    0    0    0    0    0    0    0
   16   13    7    8    1    2    3    6   11    0    0    0    0    0    0    0
   16   14   13    3    4   11    9   12   10    0    0    0    0    0    0    0
   16   15   14    9   10    7    0    0    0    0    0    0    0    0    0    0
   16   14   15   11    0    0    0    0    0    0    0    0    0    0    0    0
   16   14    1   12   11    2   13   10    7    6    0    0    0    0    0    0
   15   12   16   14    0   10    7    6   15    5    0    0    0    0    0    0

Remember that the recalls matrix shows the serial position (i.e., position in the presentation list) of each item recalled. For example, in the recall period of the first trial, the subject recalled the item presented in the 15th position, then the 16th position, then the 9th position, and so on.

Zeros indicate either no word recalled, or an intrusion (see point 1 below).

Note two important things:

  1. Intrusions (recalls that were not presented in the just-seen list) are not included in the recalls matrix above. For example, notice in the third row (i.e., third trial) of rec matrix that the last item recalled is 1591. Word 1591 was presented in trial 2, not trial 3; this is an example of a prior-list intrusion. The corresponding spot in the recalls matrix is zero, because word 1591 was not found in the 3rd row of pres.
    • For the purposes of constructing a lag CRP, we can -- and should -- ignore all intrusions, so this is ok.
  2. In-list repetitions are not noted and show up in the recalls matrix normally. Notice that in the last trial, word 87 is recalled twice, so serial position 15 appears twice in the last row of the recalls matrix. We will have to take care of that before calculating the lag CRP.

Click here for the Software page.

Personal tools
Namespaces

Variants
Actions
Public pages
Internal Wiki
Toolbox