> 1. I know you can use the <data> / <summarydata> to record additional values and such, but a few measures (latency etc.) are always measured.
> Is there a way to filter those out beforehand (as in some questionnaires I'm not interested in latency)?
The <data> element only applies to trials run via <block> elements. You can "filter out" latency by not logging any latency column per the <data> element's /columns.
. <survey> output is in a different format (wide: one column per question, only one row per participant) and cannot currently be customized, so no, you cannot suppress the logging of latency for surveypages / survey questions run via a <survey> element.
> Related to this, as in some trials I am interested, is the latency measured as miliseconds?
Yes, latency is measured in milliseconds.
> 2. Also kind of related, for one task I use <checkboxes> instead of <radiobuttons> due to the question page layout, and just assigned the option value to them.
> But in the outputfile, every single one is listed up, regardless of whether it was chosen or not.
> Is it possible to somehow filter out the unchecked <checkboxes> beforehand (since the output file gets quite long otherwise)?
No. For <checkboxes> all options are logged because -- as opposed to <radiobuttons> -- checkboxes in principle allow for selecting _multiple_ options, not only one.
> 3.This one is a bit different and not entirely related to output itself but: At one point, participants have to answer question about an image which is randomized.
> In order to know what picture they got, I used a value which gets randomized (using rand()) before the page) and is used as the /select for the pictures.
> and record it to the data. This seems to work fine, as the picture changes during different trials. I noticed though that in the output data the random value is not "whole"
> numbers but instead has decimal points, so if I used rand(1,4) to randomly choose between 4 pictures, I'd get numbers like 1,3494 and such.
> This makes it rather difficult since I don't know how the rounding works (e.g. if it rolls a 1.563, does it chose picture 1 or 2?).
> So is there a way to automatically round up?
You'll want to do
round(rand(1,4)) instead of just rand(1,4) if you want the result to be an integer.