How many missed flips is too many in Psychtoolbox (MATLAB)?

Question

I am running a reaction time experiment with Psychtoolbox in MATLAB (R2014b). I'm using SyncTests, and that gives me an estimate of how many flip commands missed the requested response deadline when I run the experiment. What number of missed flips is too high (i.e., cause for concern)? I usually get around 140 per 1900 flips, so a little over 7%. Is that okay?

Here's the message I receive (the actual numbers vary slightly after each run of the experiment, naturally): "INFO: PTB's Screen('Flip', 10) command seems to have missed the requested stimulus presentation deadline a total of 130 times out of a total of 1900 flips during this session."

Here's the code I use to flip (and then to measure RT which is the most crucial part of my code):

WaitSecs(.2); [this is the intertrial interval, more or less]
Screen('DrawTexture', mainwin, Target);
Screen('Flip', mainwin);
timeStart = GetSecs;keyIsDown=0; correct=0; rt=0;
while 1 & (GetSecs - timeStart) < 3 
      [keyIsDown, secs, keyCode] = KbCheck;
      FlushEvents('keyDown');
          if keyIsDown
                nKeys = sum(keyCode);
                if nKeys==1
                        if keyCode(Left)||keyCode(Right)||keyCode(Down)||keyCode(Up)
                            rt = 1000.*(GetSecs-timeStart);
                            keypressed=find(keyCode);
                            Screen('Flip', mainwin);
                            [etc., I just close all loops/if statements and move on
                            to the next trial]

A bit of context: I'm running my paradigm on Windows 10, and there is also a "DPI-awareness" issue I can't get around (I get a warning that MATLAB is not DPI-aware on my computer). I'm unsure what problems this might create - but the paradigm seems to be doing what I want it to be doing plus it looks great (i.e., the images are displayed nicely), so I never worried about that that much. Should I have? Can that mess with the accuracy of reaction time measurement?

Answer 1

The below suggestion is a little incomplete because I don't have the rest of the experiment, but I think you'll get the idea. You should be able to get greater precision by scheduling the screen Flip in advance. In this example the time between the previous and current stimulus is computed as frames, and the current stimulus is then scheduled for 1/2 of one frame less then you'd like it to be presented. Psychtoolbox will then present it at the next possible frame, which is when you'd like it to be presented (since you can't present a visual stimulus between frames).

mainwin = Screen('OpenWindow');

% fetch the interflip interval of the screen
ifi = Screen('GetFlipInterval', mainwin);

% inter-trial interval
ITI = .2;

% start of block, flip a screen at an arbitrary time to get a reference
% point for the first trial
Screen('FillRect', mainwin, 0);
[~, lastVisualOnset] = Screen('Flip', mainwin);

% ID the number of frames until the next stimulus should be presented
% trial-by-trial timing jitter could also be added here if desired
nextStimFrameDelta = round( ISI / ifi);

Screen('DrawTexture', mainwin, Target);

% schedule the Flip in advance by .5 frames less than needed,
% so that the next available frame will be the one wanted
% this visual onset time stamp will then be the reference point for the
% next trial
[~, lastVisualOnset] = Screen('Flip', mainwin,  lastVisualOnset + ((nextStimFrameDelta - 0.5) * ifi));

keyIsDown=0; correct=0; rt=0;
while (GetSecs - lastVisualOnset) < 3
    [keyIsDown, secs, keyCode] = KbCheck;
    FlushEvents('keyDown');
    if keyIsDown
        nKeys = sum(keyCode);
        if nKeys==1
            if keyCode(Left)||keyCode(Right)||keyCode(Down)||keyCode(Up)
                rt = 1000.*(GetSecs-lastVisualOnset);
                keypressed=find(keyCode);
                Screen('Flip', mainwin);
            end
        end
    end
end