A Likert scale looks like the simplest part of a questionnaire and is one of the easiest to get wrong. Unbalanced anchors, a missing midpoint or uneven spacing all bias your data, and none of it can be repaired once responses are collected.
Pick what you are measuring and how many points you want, and this tool produces properly balanced anchors together with the scoring rules, the analysis that scale actually supports, and the item-writing rules that decide whether the data is usable.
A balanced scale has equal numbers of positive and negative options with symmetrical wording. If one end is "strongly agree", the other must be "strongly disagree". Anchors such as Excellent, Very good, Good, Fair put three favourable options against one unfavourable, and will lift your mean regardless of what respondents actually think.
Spacing matters as well. Respondents read the options as roughly evenly spaced, and that assumption is what allows a summed scale to be treated as interval data. A scale jumping from "never" to "always" with only "sometimes" between them breaks it.
Five is the usual default: it gives a midpoint and adequate range without tiring respondents. Seven allows finer discrimination and slightly more reliable summed scores, at the cost of a longer form — sensible for an established instrument, less so for a first questionnaire.
Four points removes the midpoint and forces a direction. That can be justified if you believe respondents use the midpoint to avoid deciding, but it has a real cost: someone with genuinely no view has nowhere honest to go, and may guess or abandon the form. Whatever you choose, justify it in your methodology and keep it consistent across the instrument.
Code the options 1 to n in order. Any negatively worded item — included deliberately to catch respondents ticking straight down a column — must be reverse-coded before analysis, or it will pull against the items it is meant to support. Forgetting this is among the most common causes of poor reported reliability.
Where several items measure one construct, report Cronbach's alpha; above 0.70 is the usual minimum for internal consistency. A low alpha generally means either the items are not measuring the same thing or a reverse-coded item was missed.
A single Likert item is ordinal data. The distance from "agree" to "strongly agree" is not demonstrably equal to that from "neutral" to "agree", so a mean of one item is difficult to defend. Report the median and the frequency of each option instead.
A Likert scale — several items summed or averaged to measure one construct — is conventionally treated as interval, and that is what permits t-tests, ANOVA, correlation and regression. Distinguishing a Likert item from a Likert scale explicitly in your methodology is worth doing, because supervisors and external examiners do ask.
One idea per item. "The service is fast and affordable" cannot be answered by someone who finds it fast but expensive, and you will never learn which half they answered. Avoid leading wording, double negatives, and vocabulary your respondents may not share.
Where respondents are more comfortable in Nepali than English, translate the items and back-translate to check the meaning survived — a mistranslated anchor silently changes what you measured. Then pilot with five or six people from your target population and watch for hesitation or re-reading, both of which mark ambiguity.
Five is the common default and quicker to complete. Seven gives finer discrimination and slightly better reliability for summed scales but lengthens the form.
Usually yes — without one, respondents with no genuine view must take a side. Removing it can be justified, but state the reasoning in your methodology.
For a single item, prefer the median, since one item is ordinal. For several items summed into a scale measuring one construct, a mean is conventional and supports parametric tests.
Flipping the scores of negatively worded items so they point in the same direction as the rest. Skipping it commonly produces a misleadingly low Cronbach's alpha.
If your respondents are more comfortable in Nepali, yes — and back-translate to confirm the meaning is preserved. Note the translation process in your methodology.