A formerly small question takes on a more significant meaning in this year's salary budgeting process. The question, essentially whether to Count the Zeros or Discount the Zeros, is covered at length by JC Kovac and Jim Stoeckmann of the WorldatWork blog Compensation Conundrum
JC describes the dilemma underlying the question as follows:
A lot has been made about the 2.2% salary budget projected for 2009, and how it is a drop of 1.6 percentage points from the 3.8% of 2008. I have read many articles quoting our survey and others discussing the disturbing trend of lower increases and projecting those increases into the future — the "new normal."
Now, I am struggling with this data. The SBS is accurately reporting the average increase at 2.2% — however, that number includes all organizations that are reporting 0% increases this year. In previous years participants who projected no increase (0%) numbered only 2-3% of total respondents, but this year that number jumped significantly (30-40%+ based on employee category). By removing the 0% respondents, the average increase (of those giving an increase) sits around 3.1%. Now, 3.1% is still less then 3.8% of yesteryear — but it is still significantly better than 2.2%.
Recent advance data released by Watson Wyatt (and covered here) presents a similar set of data (and questions) as the early WorldatWork results.
For those of us seeking high level guidance on where to set next year's salary budget ... what to do? Do we use the figures with zeros or the figures without zeros?
The answer is ... Yes.
(You knew I wasn't going to make this easy, didn't you?)
Yes ... because both numbers tell you something important about what's happening in the labor market.
The "with zeros" data gives you a snapshot of the market in its entirety, the speed at which salaries are moving on average, considering all employers, those who are/have awarded increases as well as those who are not/have not.
The "without zeros" data tells you what is happening at those employers who have remained financially strong enough to commit to salary increases and who will be in the best position to compete for your employees.
How much relative weight you give to either figure must reflect your particular situation, including, as Jim says, "things like turnover, time-to-fill hiring trends, hot skill concerns, strength of business performance, and similar data related to an organization's particular circumstances."
These are interesting times, and interesting times will demand that more thought and analysis be applied to what may have been relatively straightforward decisions in the past.
No way around doing your homework this year.
Maybe its time to start reporting the numbers as a range of responses instead of a single average? For example, maybe the best way to report the numbers is like this: (using made up numbers for illustration only)
Budget Population
Change Surveyed
0%-0.9% 30%
1%-1.9% 10%
2%-2.9% 10%
3%-3.9% 30%
4%-4.9% 15%
5%-5.9% 5%
6%-6.9% 0%
Posted by: Paul Weatherhead | August 03, 2009 at 02:58 PM
Paul:
Why not both ways? I like having a single average, so I can track changes and trends more easily, but it is also helpful to see response ranges as you provide in your example.
Posted by: Ann Bares | August 03, 2009 at 03:27 PM
Agreed!
Posted by: Paul Weatherhead | August 03, 2009 at 08:43 PM
I agree that both numbers, with and without the zeros, are important. I like the idea of having the distribution. So much of the results are dependent on the industry. You've really got to do the homework this year, it's not a normal year for salary increases by any means!
Posted by: Matt N Johnson | August 03, 2009 at 10:05 PM
Matt:
It doesn't seem to be a normal year from a number of angles, does it?
Thanks for the comment and thought!
Posted by: Ann Bares | August 04, 2009 at 05:54 AM
How about using the median instead of the mean? That would take out the skew of the zeros.
Posted by: Greg | August 04, 2009 at 12:05 PM
Greg:
Interesting point. With zeros making up (according to JC and Jim) 30%-40% of the responses, though, I have to think that they are having an impact on the median as well as the mean. Unless I'm missing something ....
Posted by: Ann Bares | August 04, 2009 at 12:47 PM
No, I believe medians would not affected as are averages by the extremely low scores like zeroes. Makes no difference when you answer the question: "what is the middle number?" whether the 49% below the center are 0% or 1.8% or spread out from at every decimal level... the middle will still be the middle regardless of the size of those in the higher half or the lower half.
If there's a math PhD out there who knows different, please correct me if I'm wrong.
This is why reliability statistics are so important: to show the median, mean and mode and standard deviation; so you know what the numbers really mean. As Benjamin Disraeli famously said, "There are lies, damn lies, and statistics."
Posted by: E James (Jim) Brennan | August 13, 2009 at 09:44 PM
This post (as well as the reader responses) are very timely. Thanks!
Posted by: Devoted Compensation Professional | August 26, 2009 at 03:03 PM