There are several considerations in determining where to focus your efforts.

1. First, look for the items that rank highest in importance -- the ones with the highest correlations with overall satisfaction. Typically, their correlations will be +.50 or higher. These are the items which merit top priority.
2. Next, see how well you are rated on those items.
   • The ones on which you receive high ratings don't need improvement -- but it's important not to let performance slip, or you could be in trouble.
   • Pay close attention to those on which your ratings are not so good -- they show not only where you need to improve, but also which improvements offer the best pay-off in increased employee satisfaction.

     Different types of people often have different needs in the workplace. If sample size allows, it is important to redo the importance analysis for key segments, such as people of different ages, incomes, or job types. It's not uncommon to find that different variables drive, for instance, the satisfaction of exempt and non-exempt employees. That means any across the board shift in HR focus may leave a large number of employees unhappy. Shifts in HR focus may need to be tailored to target the specifics of different groups.

3. Items with less than +.30 correlation to overall satisfaction deserve less attention, even if the performance scores are dismal. Improved performance won't have any consequential effect on overall satisfaction.
4. Finally, always pay attention to any specific complaint mentioned by a large proportion of employees in response to open-ended questions. What's needed to be done usually is self-evident.

What is the Relationship Between Attribute Correlations to Overall Satisfaction, and Importance?

The correlations are calculated between overall satisfaction and each of the attributes and benefits. A perfect correlation would be "1," indicating that an attribute always is rated higher by the same relative amount as the overall satisfaction score for each individual. No correlation would be "0," and a perfect negative correlation would be "-1."

The attributes and benefits are sorted from high to low, based upon correlation to overall satisfaction. In effect, this yields "derived" importance. This differs from "stated" importance, in which people rate the importance of each attribute. In the opinions of many, derived importance is a superior measure. While it is quite possible for someone to "state" that something is the most important item, the derived importance may find it to be relatively low in importance, as it relates to overall satisfaction. The reverse also holds true.

By using derived importance, you can avoid focusing on improvements to attributes that will not have a positive impact on overall satisfaction. For example, an attribute may have a relatively low "performance" (agreement) score, but you wouldn't want to do much about it if it had a low correlation to overall satisfaction. The example I like to use is "carpeting." Suppose we had included an attribute named "I love the color of the carpeting." One wouldn't expect this to be as highly related to overall satisfaction as "I am treated as a valued employee" (which almost always ranks in the top ten in importance), because it has little to do with satisfaction. I would expect the correlation on such an item to be very low, which would indicate that you shouldn't run out and buy new carpeting.

Now, one caveat may be in order here. Some attributes may have ranked high in stated importance, but will be low in derived importance. This doesn't mean that you can cut back efforts in these areas. What the data are saying in this case is that "things are fine as is" with regard to that particular attribute. Letting performance slip may cause an attribute to rise in importance the next time you survey participants.

Why Ask Demographic Questions in an Employee Satisfaction Survey?

In satisfaction questionnaires, including employee surveys, demographic questions are almost always included. These questions serve two main purposes:

1. To see how closely the sample replicates the known population. The more closely the demographic distribution of survey respondents matches the population, the more confidence you can have in the data.
2. To allow analysis of sub-groups of those responding to the survey. It is this second purpose, analysis of sub-groups, which provides the most utility. If the sample size is small (less than 100 or so), this cannot be done. However, with larger sample sizes, an analysis of sub-groups can tell you things that would be missed by looking at the aggregate data. For example, you might find that 10% of employees would rate the health care plan provided by your firm as "poor";. You might be tempted to conclude that you do not have a major problem in this area. However, if you were to drill down into the data and find that 30% of employees with children rated the health care plan as "poor," you would have a potentially serious issue on your hands.

Questionnaire length usually limits the number of demographic questions you can include in your survey. The particular items you choose to include will be influenced by the type of industry you are in and the composition of your work force. Below is a list of the types of demographic questions you should consider including in your employee satisfaction questionnaire.

• Age
• Sex
• Marital status
• Number of children present in the household
• Length of time employed by the firm
• Exempt or non-exempt (or managerial or clerical, etc.)
• Department (if the sample is very large)
• Annual income (in broad ranges)

It is not necessary to include all of the above questions in the survey. The key in selecting the items to include is determining, to the extent possible, the variable that you believe will segment the population in a manner such that the different segments will have different needs and/or feelings about the organization.

Evaluating Employee Comments

Many employee satisfaction surveys give employees the opportunity to provide comments. Evaluating each comment as it is returned will be helpful in understanding the needs of individual employees. However, you'll gain even more information if you review the comments in batches and attempt to uncover patterns in the responses through "coding" them.

This doesn't have to be rocket science. If you are dealing with a relatively small number of comments (less than two hundred or so), all that is necessary is to read through the comments once or twice and write down "categories" into which most of the comments would fit. Typical categories would include items such as speed, accuracy, courtesy, pricing, product, product availability, business hours, business location, etc. You should create one list for positive comments and a second list for negative ones, including the same categories in each. After you have developed the categories, you would review the comments again, this time recording a "tick mark" in the appropriate category for each comment. People often comment about more than one thing, so it is a good practice to break each comment into parts and enter a tick mark for each portion of a comment. So, if a comment says something like "It took too long for my food to be served and the waiter was very rude," you would code it in both the speed and courtesy categories of your negative comments list.

Once you have completed "coding" the comments, you can look for patterns. You would commonly see many of the same things identified as issues in the quantitative portion of the survey report (if you have a quantitative section). Often, the comments will provide you with more specific information about how to fix problems identified in the quantitative portion of a survey. However, it is not uncommon to uncover completely new issues by coding and reviewing comments. If this occurs, consideration should be given to modifying future questionnaires to specifically ask about such issues.

If you see a lot of references to particular employees, you should consider including employee names in both the positive and the negative comment coding sheets. You could then put a tick mark in the appropriate column for each positive or negative mention of an employee. If you do this, be sure to also place a tick mark in the appropriate "issues" category as well.

If you have many comments to code, it is helpful to use a data base program or a spreadsheet. Rows should represent each respondent and columns can represent categories. This approach also makes it easy to calculate what percentage of respondents mentioned items in a particular category.

Superior Response Rates

Babbie (1975) is often quoted as recommending a 50% response rate. However, the rest of his quote is often left off: "The reader should bear in mind, however, that these are rough guides, they have no statistical basis, and a demonstrated lack of response bias is far more important than a high response rate." It is important to remember that a low response rate creates the possibility for response bias but it is not equivalent to response bias.
Babbie, E. (1975). The Practice of Social Research. Wadsworth.

Our average response rate for employee satisfaction surveys conducted online has been approximately 75% compared to 65% for the traditional "paper and pencil" methodology. It is not unusual for us to experience response rates in the 80% to 90% or higher range (we've had 100% response rates on several occasions). High tech and small firms tend to have the highest response rates.

Likert Scale

Likert (1932) developed a scaling procedure in which the scale represents a bipolar continuum. Likert-type questions typically use a five-point scale in which two ends of the continuum are balanced by a middle category. The advantage of the Likert-type scale is the variability of scores that results from using the scale. Scales based on questions with only two responses (e.g., "yes/no") tend to be less reliable than scales using questions with five response options (Lissitz, R.W., and S.B. Green, 1975, The effect of the number of scale points on reliability: A Monte Carlo approach, Journal of Applied Psychology, 60, 10-13.)

When used in summated rating scales to produce the "interval-level data" required for certain types of statistics (e.g., mean scores and correlations), the interval from one response category to another must be assumed to be equal. The use of numbers (e.g., 1 2 3 4 5) in the rating scale provides the respondent with that quality of equidistance; the labels (very poor ... very good) above the scale act as guides to help respondents understand which end of the continuum to use for high marks and which to use for low marks.

Aggregate Scores

When answering a survey, respondents expend less cognitive energy and settle for "merely satisfactory" responses rather than the most accurate ones. As Krosnick (1999) states, it is an unrealistic hope for respondents to optimize their responses across a questionnaire. When you "put all your eggs in one basket" by using a single question as the overall measure of satisfaction, you are assuming that patients will consider all relevant experiences before responding to the question. If patients have in mind only what was asked last, such as the "Extent to which staff worked together to care for you," or the most salient aspects of patients' care (e.g., "meals were cold"), "Overall care" responses will not be optimal. Responses will be weighted certainly, but each set of weights will be biased and not the result of patients' considering the entire range of their care.

Using section mean scores to compute overall scores reduces the survey error associated with single overall questions and provides a more comprehensive measure of patient experiences. We average section mean scores instead of all questions so that dimensions with more questions (e.g., "Tests and Treatments") are not weighted more heavily in overall scores than dimensions with fewer questions (e.g., "Nurses"). In the strictest sense, each section is weighted equally in patients' and facilities' overall scores. In reality, global issues such as nursing care, communication, and sensitivity cross section boundaries and influence responses in more than one section. Other issues, such as discharge and meals questions, are more circumscribed. As a result, meals questions, for example, generally fall in the lower half of correlations between individual questions and overall scores while Nursing and Physician issues generally fall in the upper half. Whether these items are priorities for individual facilities, nursing units, specialties, etc. is determined by the unique combination of performance (as measured by each question's mean score) and relative importance (as measured by correlation coefficients) in each client's Internal Priority Index.

Handling Mistakes

Successful health care organizations strive for excellence in customer service. Despite the good intentions and best efforts of the staff, sometimes patients and family member have less than desirable experiences. Whether it's a delayed appointment, inability to find parking, a misunderstanding with an employee, lost test results, confusing directions, or worse - things don't always go the way they should, and the organization creates a hiccup in its usually stellar service. For the purposes of this article, we'll call this moment an "oops."

Health care providers are familiar with the reasons for creating the highest levels of patient satisfaction and the potential for increased patient loyalty. Nevertheless, even at the highest levels of satisfaction, most customers will only tell a few others about these positive experiences. It's possible to design effective programs to increase positive word of mouth, but most health care organizations are not yet maximizing the potential of this strategy.

Although few in number, considerably more effort is directed at highly dissatisfied customers. When we examine the Press Ganey national inpatient database, we find the incidence of highly dissatisfying experiences is relatively small. In the top 10% of the database, on average only 1% of the questions receive the lowest response (very poor, or "1" on a five point scale). Even in the bottom 10% the average of very poor responses is only 4%.

However, these dissatisfied patients have a propensity to tell many others about their bad experience. Most of us are familiar with a customer service experience so dissatisfying that we've vowed to tell "everyone we know" about it. Leebov (1993) estimates that people will repeat a dissatisfying health care story 20 times; studies in other industries estimate that on average 9-10 others are told. (TARP, 1979).

It's important to note that these studies pre-date personal computers and the Internet, and only measured negative "word of mouth." Today with computer access, it's possible to imagine a multiplier of ten or more, factoring in the potential for negative "word of mouse." The recent experience of a computer executive at a Texas hotel made its way around the world several times in a PowerPoint presentation over the Internet and eventually into the Wall Street Journal.