Welcome to the Listening Post!

Appropriate scale length depends on the extent respondents can differentiate different attribute levels in the stimuli they are rating. So: for concepts tests in which respondents are being exposed to something new, a 5- or 7-point scale will be best. For a customer satisfaction study, where respondents are rating products with which they are familiar, then an 11-point scale (0 to 10) is best. In all cases, use a balanced scale. If you go to the “bible” of psychometric theory—Jum Nunnally’s Psychometric Theory—scales measure the amount of attributes. Consequently, they have to have a midpoint.

I am not the greatest writer but I do alot of Marketing research. I recently had to convince "management" about switching from a 5 to 7 point scale. I found an simple article from Quirks and though it would be Relevent to this discussion. 
 
basically there is no real conclusive evidence to suggest a one scale system. the question know becomes unipolar or bipolar- It seems Maritz uses the unipolar more 
 
 
 
Balanced scales have equal numbers of positive and negative points: completely satisfied; mostly satisfied; mostly dissatisfied; completely dissatisfied. Although there are exceptions, a best practice with respect to balanced scales is that the use of modifiers should be symmetrical on the positive and negative ends of the scale. 
 
 
 
Unbalanced scales attempt to get greater discrimination on one side of the scale than on the other. If past experience suggested that most respondents are satisfied with a certain product or service, a researcher might want to “stretch out” the positive side of the scale, as in this extreme example (five positives and only two negatives): completely satisfied; very satisfied; mostly satisfied; somewhat satisfied; barely satisfied; mostly dissatisfied; completely dissatisfied. 
 
 
 
Whether a scale should be balanced or unbalanced usually depends on whether we’re measuring a unipolar or a bipolar concept of satisfaction. A unipolar satisfaction scale might range from “not satisfied” to “completely satisfied” (i.e., it doesn’t measure any more extreme dissatisfaction at all). In contrast, a bipolar scale would range from “completely dissatisfied” to “completely satisfied” (i.e., it measures extremes of both satisfaction and dissatisfaction). 
 
 
 
When we don’t know ahead of time whether most respondents will tend to be satisfied or dissatisfied, or if we expect high levels of dissatisfaction, a balanced bipolar scale is appropriate. If we know from past experience to expect low levels of dissatisfaction, an unbalanced unipolar scale will be better. In our firm’s experience, most studies show low levels of dissatisfaction, so our most frequent scale recommendation is for an unbalanced satisfaction scale. 
 
 
 
Some people think that satisfaction and dissatisfaction should be measured separately and are different entities. Although Maritz has observed that satisfaction and dissatisfaction may sometimes have different drivers, and there may be nonlinear relationships between satisfaction and other variables, we have seen no compelling reason to measure them separately. 
 
 
 
C. Midpoints 
 
 
 
A midpoint communicates neutrality on a balanced scale. “Neither satisfied nor dissatisfied” serves as the midpoint in this balanced bipolar scale: completely satisfied; mostly satisfied; neither satisfied nor dissatisfied; mostly dissatisfied; completely dissatisfied. 
 
 
 
There seems to be no difference in quality between scales that have a midpoint and those that do not. For bipolar scales, however, Maritz Research follows the advice of Sudman and Bradburn (1982) to “include the middle category unless there are persuasive reasons not to do so.” Krosnick (in press) suggests the same. 
 
 
 
D. Number of scale points 
 
 
 
Despite the strong opinions of some scale enthusiasts, there just isn’t powerful empirical evidence that a single number of scale points are always the best. A literature review by Cox (1980) summarizes the consensus regarding self-administered surveys: “Seven, plus or minus two, appears to be a reasonable range for the optimal number of response alternatives.” Internal research has not indicated a consistent difference in quality between 5- and 10-point scales. Krosnick (in press) finds that 5-point scales have the greatest test-retest reliability in paper-and-pencil surveys. Maritz’ experience with mail studies has shown that 5-point scales provide better dispersion of responses (and are, therefore, more discriminating) than other scales. We expect these results would generalize to other visual survey modes (e.g., PC or Web-based). 
 
 
 
The findings about the superiority of 5-point scales apply specifically to visually perceived scales and may or may not generalize to questions that respondents hear rather than see. Ten-point response scales are common in telephone surveys and have also been found to provide some advantages: greater statistical precision, statistical power, and room for improvement (Wittink and Bayer, 1994). We have also seen evidence that respondents from some ethnic groups shy away from using the extreme endpoints and may answer more accurately if they have a greater number of scale points from which to choose. Hence, while our basic recommendation is to use 5-point scales, we are comfortable using 10-point scales for some surveys, particularly those conducted via telephone. 
 
 
 

For a seven point bi-polar scale is there consistent way to label the points?  
 
 
 
Extremely Satisfied 
 
Satisfied 
 
Somewhat Satisfied 
 
Neutral 
 
Somewhat Dissatisfied 
 
Dissatisfied 
 
Extremely Dissatisfied

That certainly works, though I prefer to start survey scales with the negative choice first. The 7-point satisfaction scale I use: 
- Completely dissatisfied 
- Mostly dissatisfied 
- Somewhat dissatisfied 
- Neither satisfied nor dissatisfied 
- Somewhat satisfied 
- Mostly satisfied 
-Completely satisfied 
You can label the midpoint “Neutral” for succinctness, if you prefer. 

Question: I am a graduate student conducting a survey to determine worksite safety culture from a construction workers perspective. I intend to use scaled questions in my survey. Example questions are the following: 
 
 
 
Please rate your level of agreement to the following statement: 
 
1. My coworkers support the company's safety program 
 
2. Employees correct other employees unsafe acts 
 
 
 
I considered doing a 5 point scale to measure agreement, but then considered using a 4 point balanced scale. My thinking is that either workers support the safety program or they don't. There is no "neither." As my sister would say, "Either you are pregnant, or you are not." 
 
 
 
I would appreciate any advice regarding this.

Typically agreement scales are bipolar (disagree to agree), and the best bipolar scales have seven items. I'd use: Completely disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Completely agree 
 
Many respondents select "Neither agree nor disagree" (sometimes "Neutral" is substituted) as a "No opinion" measure.

On the whole, we use 10-point scales, labeling both end-points, with 1 = lowest measure (i.e., Does not describe me at all; Not at all important; poor, etc.) 5 - 6 labeled as the neutral measure (i.e., Describes me somewhat; somewhat important; average/acceptable, etc.) and 10= highest measure (Describes me completely; extremely important; excellent, etc.).  
 
When interpreting results, we typically look at means as well as Top-4 Box scores. We have found that unless we define the midpoints, we risk mis-interpreting the 7 - 10 scores, where some participants are using a grading system model when selecting their score and "7" is close to "failing" vs. the low end of extremely positive. 

The 7+/-2 cognitive limit hides the fact that cognition is logarithmic. three is the baseline. At 5, you've selected or filtered another population. At 12, your population has gotten very small. Some people will not see the difficulties at 12, but the folks you lose moving from 3 to 4 will be more than 4x away from 12.  
 
Keep in mind that a logarithmic scale of base 3 puts item 4 at the linear location of 6, item 5 at 9. Twelve is way the hell out there for the 3 crowd.  
 
This has nothing to do with discrimination of the results. It's more like blowing away your statistical significance.  
 
Each cognitive population would need to be treated as its own entity, rather than as a contributor to the aggregate population. Aggregates falsify.

No one has mentioned that agree-disagree items are prone to acquiescence bias (though Jeffrey hints at this).  
 
Notice also that in the original blog post, Krosnick's response is the only one citing scientific evidence for his recommendation.  
 
Read the literature.

You are right, Matt – we should read the literature. Unfortunately, for many researchers, rating scale best practices are a matter of faith rather than science, a matter of tradition rather than best measurement.  
 
And I would encourage people to rewrite agreement scale questions; that’s another tradition that should be retired.

I appreciated your article because it provides info about when a scale is useful rather forcing the point of which is best. 
 
Once one has decided on the number of points, the next question is how label the points. From my work with my mentor Kerm Carlson (M.A in media research) who cited Likert's work, the 5 point scale provides more value distinction than a 3 point and is more intuitive for quick emotional accuracy than a 7 or 10 point. He avoided naming the points because it moved the respondent from intuitive/gut to intellect/analytical. He allowed for mid-point responses too, which gave respondents the option to express the accuracy they want (more democratic and therefore respectful). 
 
For example...evaluating "How do you feel about banana in a peanut butter sandwich?"  
 
Repelled Attracted 
1--2--3--4--5 
 
vs 
 
1 extremely repelled 
2 repelled 
3 neither repelled nor attracted 
4 attracted 
5 extremely attracted 
 
I suggest that the detailed words are unnecessary, distracting and at worse misleading when not congruent with the key criterion in the question. However, an indicator of the high and low end is essential. I use 1 for low and 5 for high so it is easy to do math and charting with the results.  
 
Underlying all this are the questions...Does my survey measure what I really want to know? and Will the answers provide info I can take action on? In my experience, many surveys fail this test. 
 
Rhetorical Survey: 
To what degree were my comments helpful? 
Unhelpful Helpful 
1--2--3--4--5