This removed the warning: "sudo apt-get install gtk2-engines-pixbuf"
But the problem did not go away. ;-(

Can you attach a screenshot?

On Ubuntu Unity, the main menu is at the top left of your screen (not at the top of the window)... Perhaps this is the point.

Err... my excuse is that I am new to Unity (so far I do not like it that much, but I guess there is more to discover that makes my impression of over-simplification hopefully go away).
So you got it: The problem is no problem. :-))

By the way, my intention is to try out AMC for secret election ballot counting at the general assembly of a large NGO, using one or two Fujitsu S1500 (on Windows) and AMC in VirtualBox. There are about 5000 ballots of A6 size that need counting, 500 in 10 elections with rules like "select up to 4 out of 6" or "select 1 out of 3". If you have any suggestions I am all ears!

1. When exporting, ask for ticked boxes list and post-process from the CSV exported file
2. Use a recent version of AMC: ticked boxes data can be found in SQLite tables

SQLite would be great - I guess Ubuntu 11.10 (svn 550) does not have that "recent version", does it?
By the way, would be nice if this would work:

> auto-multiple-choice --version

What I am most concerned with: What kind of error rate must I expect, and is there a way that AMC automatically flags the scans it had trouble with and wants a human being to take another look at?

Hm... can't edit my post it seems, so here comes another one:

Combining the elections all on 1 paper is not an option: That's a thought to consider indeed, yet it has the potential to kill the "secret" part in "secret elections" because the decisions are now connected and allow to, for example, determine a preference for candidates from a certain country.

Philipp von Klitzing wrote:

SQLite would be great - I guess Ubuntu 11.10 (svn 550) does not have that "recent version", does it?

No. You can use a version from the test PPA (ppa:alexis.bienvenue/test)... There is regular updates there, but I think it is currently quite usable.

By the way, would be nice if this would work: auto-multiple-choice --version

At present, you can only use Help->About from the menu.

What I am most concerned with: What kind of error rate must I expect

Please make tests (for example with provided sample questionnaires), as I'm afraid this depends on your hardware (printer/pencils/scanner)...

and is there a way that AMC automatically flags the scans it had trouble with and wants a human being to take another look at?

Yes. There is such an indicator to review only difficult scans.

I am not fully through the documentation, yet: Is it possible to create fully anonymous paper ballots that do not have any kind of serial number (unique identifier) on them? A common identifier is fine, actually desired, but these ballots need to be 100% identical.

Now I am smarter, after having fully finished the (well-written) documentation. While reading I felt that chapter 4 and 5 should swap their order.

So, for my non-identifiable paper ballots it looks like I would either need
a) a 500 pages MCQ (and still not be happy) for having each ballot on a new page, or
b) a config switch to generate a fake unique identifier when processing the scans

If b) would indeed work then I'd like to suggest it as a feature request; it would open up AMC for OMR processing of all kind of secret votes and secret elections.

Philipp von Klitzing wrote:

So, for my non-identifiable paper ballots it looks like I would either need
[...]

I'm sorry I did not tell about it before. Photocopying answer sheets is not possible for old versions of AMC, but is possible for new versions... You generate one answer sheet, print it, and photocopy it 100 times if you need so. Drawback: the answer sheet must have only one page, and it can be difficult to be sure you scanned each ballot once and only once if you don't scan all ballots in one session or if you have problems with the feeder while scanning...

The other possible way to do it is: make ballots with different id codes on them, place them in a basket, and every voter take one at random. It is not perfect, though.

Combining the elections all on 1 paper is not an option

• What is the practical situation? If voters put their ballots in envelopes, how long does it take to open an envelope, read the content (a few boxes ticked among 6 boxes) and write it on a paper? How long does it take to open the envelopes and scan the A6 paper? I'm not sure it is faster with the scan.
• Computer-driven data capture and results computation is not verifiable by humans, whereas reading and writing the content on a page with pencils can be checked by several people...

I'm sorry I did not tell about it before. Photocopying answer sheets is not possible for old versions of AMC, but is possible for new versions... You generate one answer sheet, print it, and photocopy it 100 times if you need so.

That's great, also because copying is usually cheaper than printing! Is there a special setting required for this?

Drawback: the answer sheet must have only one page, and it can be difficult to be sure you scanned each ballot once and only once if you don't scan all ballots in one session or if you have problems with the feeder while scanning...

Indeed, and here I simply lack the experience of working with fast 20ppm or 30ppm scanners. One of the options I am looking at is to number the ballot papers after the ballot box has been opened, right before the counting, either manually or with a rotating stamp. It slows things down, though, during numbering and especially when digging through the scans to find the duplicate one. A scanner with a built-in imprinter for the back side is a bit too expensive and would probably not help much - because printing is on the back side. ;-)

The other possible way to do it is: make ballots with different id codes on them, place them in a basket, and every voter take one at random. It is not perfect, though.

That I already had (and still have) as an option on my mind: There are no voting booths, instead voting happens in the open in a large plenary, and ballots are on the table for at least half an hour or so, i.e. the IDs are in principle good to see. Maybe I would need to instruct the voters to fold the ID away on a dedicated folding line as the first thing they should do.

• What is the practical situation?

The OMR approach should be considerably faster (7 elections with 3500 ballots can currently take about 5.5 hours of manual counting, with many persons that need to be involved), and time is in short supply because we are close to the end of the assembly. There are no envelopes, the ballots are folded in half; counting takes at least 3 persons, and tight results need to be counted twice or even three times.

• Computer-driven data capture and results computation is not verifiable by humans, whereas reading and writing the content on a page with pencils can be checked by several people...

Indeed - that is why I want to stick with paper and not do pure computer-assisted (aka electronic) voting. There is always the paper trail available for manual verification and re-counting. The first round of counting will be done with OMR, the optional second round manually.

Preparing the A6 sheets and then getting 4 of these onto an A4 sheet for printing/copying (and later cutting) appears to be more complicated than I had anticipated - tips are welcome; most likely I need to have a file I can give to a printing service and cannot print directly (in case AMC's unique IDs are employed).

Is there a size-change limit that must be observed, for example can AMC process scans that are 10% smaller than the original layout, or have 50% or even 25% of the original size?

Is there a special setting required for this?

AMC asks you if this is the case before the first data capture.

Preparing the A6 sheets and then getting 4 of these onto an A4 sheet for printing/copying (and later cutting)

Just make an A6 answer sheet (see attached LaTeX source example). The evince viewer allows you to print it 4 pages on an A4 one very simply (and you can choose to print to a PDF file).

Is there a size-change limit that must be observed

The four corner marks (circles) on the scan must have more or less the right size in relation to the page size. The "more or less" is an option in AMC - default value is more or less 20%. So if you scan the A6 page in an A4 scanner, the page covers a quarter of the page, and the circles size is 50% smaller than the expected size: this is not good, prefer a real A6 scan.

Thank you, that's all very helpful!

The evince viewer allows you to print it 4 pages on an A4 one very simply (and you can choose to print to a PDF file).

The layout is not always perfect with evince, however. You can also consider using pdfjam (or pdfnup) from the pdfjam debian package:

pdfjam --a4paper --nup 2x2 --suffix 2x2 DOC-sujet.pdf