|Perl Monk, Perl Meditation|
The most important things I've learned about performance appraisals over the years is that they're usually inaccurate, and that the inaccuracy is usually because they are too infrequent. People like to tie one annual performance review to one annual raise cycle, and the reviewer usually works on the review from recent memory. A year ago isn't recent, and six months ago is only slightly more in the "recent" category.
I put the bulk of my node in a readmore tag for the tl;dr crowd.
In a project shop, every project is a success or a failure. Every person involved did acceptably, unacceptably, exceptional, or abysmal work on that project just finished. Record data about the quality of work for that project. Collect all the data for the year to get a more objective and accurate idea what a person did for the year.
In an environment that isn't a project shop and involves just doing some type of work not assigned primarily to large individual projects, it's still possible to write a few notes about the quality and quantity of work someone does on a more frequent basis. Employers are often quick to write up infraction reports for too many days missed, insubordination, or in a 9-5 or office of 7-3 line work setting for tardiness, too. They could have a few notes about each employee every week or month for both the good and the bad to better evaluate the worth of an employee's relationship with the company.
Of course, having great status reports all year won't make up for punching the CEO in the face. However, short of any exceptionally bad individual actions, I think people are more fairly represented by concrete documentation taken over the entire reviewed period than by just whichever couple of weeks end up being before their review session. I think the people being reviewed find the process more fair and respect the results more, too. It's important if you don't want to actually hurt employee productivity and morale after the reviews that the people whose work, name, and pay are involved that they can be given a fair shake.
Anyone's opinion of someone's performance over an entire year is going to be unreliable. It doesn't matter if it's the HR staff, the immediate supervisor, the coworkers, or the worker. It's also going to be useless to try to point out inaccuracies about what transpired months ago. If you're going to review people's work, do so for the project or work term just completed. Give the worker a chance to point out inaccuracies or mitigating circumstances at that time, and include that in the report. Get the coworkers to give credit where it's due in the project wrap meeting (you do those, right?) or in your status meetings, and make sure those statements are noted. Even a supervisor doesn't know who did what part of which job much of the time and gives credit to the team as a whole or to who they expected to do the best work in the first place.
Praise exceptional work now, and address the problems uncovered now. Document all of that now, in permanent documentation. These reports should be the basis of reviews at the end of the reviewed period. This creates more accurate, more verifiable, and more objective reviews. It also provides a better system of feedback and more understanding between the employees and management. How often have you heard that someone would have done better on a review if only they had been told there was a problem with some aspect of their work?
If all of this sounds a bit like a school grading system, with periodic progress reports and discussions about how to improve one's marks, then that's no coincidence. Schools are responsible for teaching (and often training, about which there are many arguments), evaluating, and motivating people to do the work. The frequent feedback loop and targeting of specific strengths and weaknesses for students are a strong basis for getting an acceptable level of work out of as large a body of students as possible.
If you're a small enough shop that your employees are all already in constant contact and have good enough communication established to know where everyone stands, then you probably don't need employee reviews. A partnership or a master with one or two apprentices probably doesn't need to have a record of everything for everyone internally. It's just busywork to write up formal reports about employee production if it's just busywork to formally document your policies or your methods of working on a project.
If you're big enough to need a report, though, then you need to have reports often. Schools, armies, departments of transportation, and baseball fans don't usually take "pass" or "fail" as an answer. A school needs to know which students got which marks in which terms in which subjects. An army needs to know in which actions a soldier was involved and what he or she did in each. A department of transportation needs to know which lanes of which segment of which road needs what type of repair. Baseball fans want to know batting average, slugging percentage, fielding percentage, OBP, and a hole slew of other stats that would bore non-fans, and they're just the fans. Think about the coaches.
Paper, document storage, and document retrieval are no longer an excuse for impropriety on the part of employers. Put it on disk. Make sure it's indexed, cross referenced, fully searchable, and safely backed up. It's not that much work to gather the data when it's done little by little on a regular basis. It is nearly nothing to maintain compared to other data your company keeps. Protecting your workers (and their productivity) from lazy, incompetent, or vindictive supervisors and protecting management from poorly performing and litigious employees is worth the hassle.
Don't objectify your workers, but do objectify their performance. Put adjectives and adverbs to it, and possibly put numbers to it. They get numbers in response to an evaluation: retained or not, some percentage of a pay raise or cut, or some bonus or amount of probationary period (even if all of the listed turn out to be zero). That number seems a lot less arbitrary and capricious if there are good data to back it up. More importantly, it really is less arbitrary and capricious. In case you haven't guessed, making arbitrary and capricious decisions which directly impact employee's pride and paychecks is not so good for morale.