I think it's arguable whether eliminating the midpoint will reduce social desirability bias. I can see someone feeling forced to take the socially desirable response rather being in the position of opposing it (thus increasing the bias), while having a midpoint available would provide an alternative option. As to whether it is desirable to eliminate the midpoint, that depends on the research question and the nature of the data. In general, people often do have no opinion, and it seems most realistic to measure that, and not to frustrate people by failing to give an option that they would want. But maybe you are thinking of a different type of 4-point scale rather than a Likert scale without a midpoint.
I concur with the preceding answers. I believe a midpoint is beneficial to ensure that respondents may better provide truthful, more accurate responses.
In my opinion.. midpoint (such as neutral, undecided, neither agree nor disagree) is best used if you want to provide middle-category for moderate respondents.
in case for not using midpoint.. i think that would be somehow "forcing" the respondents to choose (either agree or disagree, for example). what if the respondents is actually moderate and does not have the answers that represent their condition?
for reducing social desirability bias, i guess we can use some filler items like the Life Orientation Test-Revised (LOT-R) (Scheier, Carver, & Bridges, 1994) or using the Marlowe-Crowne Social Desirability Scale.
i agree with the above observations. indeed my bank has moved even further and now provides a ten-point scale to measure customer sanctification with their services. they argue that this reduces bias compared to the 5-point scale. as a customer, you have more options to choose from. maybe Mr. Camilleri can look at the merits and demerits of increasing or decreasing the scale.
The world is full of Likert scales. They are everywhere; the real issue is not wether you should use even or odd point scales, but if the Likert scale it's the best option in your toolbox. Social desiderability impacts every single survey, but not always with the same strength. It's a matter of choice: the reasearcher should try to evaluate how much will social desiderability influence the results and the impact of forcing the respondants to "take sides" on a topic. Likert scales have a long history of biasing results by "creating" opinions out of nothing (it happens, when the respondants have never thought about the issues they get questioned about).
Anyways, answer "3" on a 5-point scale isn't always the most socially desiderable option. Context it's essential. So, there's no point in generalizing.
Do you agree that public administrations should avoid violence towards women? SA, A, NAND, D, SD. Obviously, social desirability does not lie in the NAND answer!
when you measure the attitude using Likert scale one should follow odd number pattern of scales like 3 point scale or 5 point scale or 7 point scale.
never use even number scale points like 4 or 6 or 8. even number scale points don't cover the entire range of the what you are going to measure, it lead to biasness
I agree on the even number with an additional "no opinion" option that can be analyzed to "equal, both, neutral, or neither", or even a comment line that can be subject to content analysis. Often the midpoint represents (and obscures) a variety of different items to study participants, so it may be worth making these options explicit and quantifiable.
I have always used 5-point Likert Scales. I believe that the respondents should be allowed to choose 3 (in a 5-point likert scale - when they are neutral).
I asked this question as I came across a paper that has presented an argument in favour of even-number scales. Like many of you, I also strongly disagree with this practice.