Know-How Exchange : Research/Metrics

It depends on what your objectives are and if and how you are going to act on the results.
If some major business initiative is planned based on the results you could think of taking " excellent and very good" answers as the only agreement ones.
I would also think how reframing the question would affect the results.
What about open-ended question instead of a close-ended one?
Are you sure everybody will have an opinion on the issue? I do not find " I do not know or have no opinion" among options for respondents.

The most objective results will be reached by qualifying the responses as a 'level' of agreement rather than a 'rate', and then present them as a pie chart, with each response having a percentage based piece of the pie.

Make sure your questions are formulated in such a way that you understand WHY the respondents attribute the standards of excellence or just good, and that you can identify what needs to better satisfy the respondents who only answered with 'good ' and 'very good'.

Any objective measurement should also take in account attribution standards such as 'neutral', 'bad' and 'very bad'.

I agree with Lorenz, in that if you're measuring agreement, the scale should run from "Completely disagree" to "Strongly agree." As he mentioned, it's important to include a follow-up open-ended questions to better understand why respondents gave the ratings they did.

For example, for those who said it was very good, ask why, in particular, do they rate it as very good? Same for those who rate it poor or very poor.

If you're measuring experience with a product or service attribute and are aiming for a linear 5-point scale, your labels are off. To measure the full spectrum of potential ratings, it should be:

1= Very poor
2= Poor
3= Neutral (no feelings either way - could also be "neither poor nor good")
4= Good
5= Very good (or excellent)

To analyze, in addition to calculating the means for comparison, you can look at the percentage of respondents in the top two and bottom two box scores.

If your sample is large enough, you can slice the data by key demographics or other segments to see where there may be differences in the ratings.

It seems like rather than looking for an agreement rate, you are looking for a satisfaction rate. Typically, this is calculated on what is called a "top 2 box" basis, where the number of excellent and very good responses are divided by the base answering the question. It can also be top box (excellent only) or top 3 box (excellent, very good and good) - you just need to be consistent and clear in your definition.

Hope this helps.

It sounds like you're just trying to summarise the data to make it more easily digestible. If it's a single study with no comparison to external data, then so long as you are consistent you can choose. However, for fairness for the reader/viewer it is normal to chart all the data so readers can see where the numbers came from, even if you only highlight key parts of the data (use color and labelling carefully for this effect). I recommend you read Tufte.

If you are trying to match external data then you naturally have to follow what ever aggregation scheme they use.

If you are trying to link scale-based data to actions (eg to likely purchase agreement) then you will have to take a guess on the real conversion rate. Most conversation guesses estimate around 60-70% of top two items turn into sales - but they are guesses and ignore promotion and distribution effects.

If you are looking at purchase estimation, consider techniques like conjoint or choice analysis. Scales are notoriously wishy-washy in what they actually mean.