Logging errors in contests can exact a steep penalty. Not only will the sponsor remove QSOs containing errors they also assess penalties. On average a removed QSO will reduce your score by at least double that amount. If the removed QSOs are multipliers the cost can become very expensive indeed.
With the advent of computer logging, log submission and log checking the wages of sin will always be paid. In the olden days logs had to be manually checked. This was such a labourious process that logs with scores not high within their entry classes were rarely checked. Carelessness that was once penalty-free for 90% of contesters has evolved and is now a guarantee of score reduction. This is not only embarrassing it can also keep you out of contention for awards.
Is it possible to submit an error-free log? Yes, but it takes a combination of diligence and luck. I'll show you what I mean by luck in a moment, but let's first cover the errors that diligence can prevent. Computer logging is both a blessing and a curse in this regard. Software makes it easy to avoid some mistakes and easier to make others.
Call sign and exchange errors
These are the easiest errors to make and often can be completely eliminated by taking appropriate care. Errors include copying mistakes, typing mistakes and pre-fill oversights. Here are some of the ones I am most familiar with, along with ideas on how reduce mistakes.
Busted calls: This can happen on both SSB and CW. In the former case it is usually due to QRM or difficulty understanding a non-native English speaker. In the latter case the causes of copying errors are more numerous. A sequence of short letters (e.g. E, I, T) or characters with 3 or 4 dot sequences (e.g. B, V, 4, 6) at high speeds will often do it. Although there are some contesters with poor CW skills who get by due the short exchanges, everyone is at risk.
If you're running the other station will most often correct your error since you are sending their call to them. However if you're sending at 30 wpm or faster the other guy may not notice the error. I'm surprised how often I hear this happening while waiting to call someone. Worse, they may not care since in the majority of contests they are not penalized for the other guy's copying errors!
Some stations while in S & P mode make a point of sending the running station's call just to be sure they've correctly copied it. But this can be expensive in time if it is done on every QSO. I would only recommend doing this when there is some doubt about the other's call.
Some logging software will flag calls that are suspect by dint of not being known as having appeared in previous contests -- a so-called master database. Flagged calls should be double-checked, just in case. Be particularly wary of partial call databases, used by you or the other station, that will attempt to pick a likely call from the master database if the call is suspect or incomplete. It is often wrong. I am a regular victim of this feature since the other station often has difficulty copying my QRP signal.
Exchanges: In most contests the exchange is fixed and often predictable from the call. For example, in this past weekend's ARRL DX CW contest my exchange was "599 ON". The RST is of course routinely sent as 599 in every contest exchange. The state/province is a constant and in the case of Canadian calls is strongly correlated with the call sign prefix. Most logging software will analyze the call and pre-fill the exchange field for you, saving you time and potential for a typo. This can also be done for CQ and ITU zones in contests where those are part of the exchange.
US locations in particular are less predictable. It has been many years since a call sign prefix correlated to a state or zone. Logging software makes a best guess but it is up to you to confirm that what the other station sends is what the software pre-fills. It is too easy to get lazy, especially when you're tired, and not confirm the fill before logging the QSO. I can only say that you must be diligent. This is the cause of some of my errors.
If you are uncertain of your copy in contests with a more complex exchange (e.g. Sweepstakes) you should immediately ask the other station for confirmation or a repeat.
Some logging software allows you to use a history file built from previous contest logs (your own or by importing one) to more accurately pre-fill the exchange. The percentage of mistakes will be reduced, but you must still be diligent and confirm what is sent matches what is pre-filled.
Correcting errors on the fly: When I note an error and make a correction on the fly it is not always possible to record the correction in the log. Even if you're highly effective at manipulating the logging software this can take several seconds, during which the other station is wondering what's going on while others waiting for you (if you're running) may QSY. I keep a pen and paper handy to jot down the needed correction since it is sometimes the fastest way to move onward. The log can be updated even just a minute later while your memory keyer or DVM is playing a CQ.
Another technique during S & P is to stick around a few seconds to hear for a second time their call and/or exchange. The call can be quickly corrected this way. However there is a danger if the error is in the exchange. If the running station has a poor rate you could be waiting a while for their next QSO to hear the needed information. Rather than getting into this undesirable situation you should not be shy about asking for a repeat during the QSO.
Not-in-log errors (NIL)
NIL errors are unlike the errors discussed above in that they are not entirely under your control. That is, you cannot be certain that the other operator logged your QSO. You can reduce NIL errors but not eliminate them entirely. This is a topic I previously discussed so you can reference that article for background. What I want to address now is some of the "why" NIL errors are so difficult to eliminate.
I will use a real example. Although the results are not out I am one of the participants in last fall's (November 2014) CQWW CW contest sent a preview log check report (LCR) for review. My error rate is comfortably under 1% so there is little risk of losing my provisional #1 North America position in the SOAB (single op, all band, unassisted) QRP category.
Apart from a small number of busted calls and improperly recorded exchanges there were 3 NIL errors. In trying to understand what might have happened, despite my effort to eradicate NIL during the contest, Randy K5ZD suggested I look through the public logs. He seemed sufficiently confident in their log checking software and processes to punt the question back at me. So I did what he suggested.
Case #1: This occurred while I was running on 15 meters. Getting an NIL while running is odd since the other station would seem to have obviously copied me or would not have called. I have assumed, perhaps mistakenly, that the majority of NIL errors occur while S & P. This is an ongoing risk to those of us operating QRP, and especially so on low bands. Unfortunately the other station's log in this case is uninformative. He was indeed on 15 at the time but his log does not provide the frequency for each QSO, just the band (21000).
Case #2: This one was also while I was running, this time on 20 meters. The other station was multi-single so the log contains the interleaved QSOs of two operating positions. One of those was on 20 at the right time, however that station was running and positioned far from my frequency.
Case #3: Unlike the previous two cases this is one that is easily determined. It was a European on 80 where all of my DX QSOs are marginal. His log confirms he was there and running, and I have some recollection of the QSO. I recall that he did copy my call, with difficulty, though he seemed uncertain and may not have copied my exchange. After a while he moved on, though he first gave me some indication that the QSO was complete. Usually this is a "TU" or something similar before soliciting the next QSO, although I don't remember what it was in this case that decided the case for me that I should log the QSO. I may have been unduly rash since it was a multiplier, something that is very precious to me on 80!
Resolution: As Randy noted confusion often reigns during a contest and some NIL of the sort I had are not uncommon. It is entirely possible for an adjacent QSO (of which you might only hear only one side) can have just the right timing to make it seem that it was with you. This may have occurred in case #1. I have no idea what might have happened in case #2. The call was that of a big gun and not likely to be a copying error on my part, and even if so the log checking software would most likely have determined the correct call from scanning other submitted logs. On this basis there is little I could have done to avoid these two NIL errors.
Case #3 is a dilemma. When there is uncertainty in a situation like this there is no best choice. If I log the QSO and it is not logged by the other station I lose the QSO and multiplier and get assessed a further penalty. If I don't log it not only I will lose the QSO and multiplier, which the other station might have logged, I could cause the other station to suffer an NIL. That would be inconsiderate of me.
This comes back to suggestions above regarding getting as much confirmation of critical data during the QSO to eliminate these doubts about whether to log it. On low bands where the other station is barely able to copy me this is difficult. That does not rate as a justifiable excuse. You just do the best you can.
Getting to a 0% error rate is difficult, as I hope my experience has shown. It is not purely a matter of chance: if you are sloppy I can guarantee you will never achieve 0% errors. Yet even if you are extremely diligent I cannot guarantee you will eliminate all errors. The reason is that what the other operators do, the hundreds or even thousands of them you work in a contest weekend, is out of your control.
At its fundamentals the error rate is a stochastic process, where there are one or more random variables. We can draw an analogy from the game of golf. The top players can never guarantee getting a hole-in-one, not in any particular match and perhaps not in their lifetimes. What they have is a higher probability of doing so than everyone else who has ever picked up a golf club. This is because they can more often hit their tee shots onto the green. If you can't do that, or do it often enough, you will never score a hole-in-one. Unless you get very, very lucky.
Operating a contest is much the same. You may never score a 0% error rate but your chance of doing so dramatically rises when you make the effort to reduce errors. Even if you don't get to 0% you will see a big improvement in the difference between your claimed and published scores, and that will push you up the leader board past those who don't make the effort.