What is Item–Total Correlation?
Item–total correlation measures how well a single survey question (an item) relates to the overall score for the scale it belongs to. It shows whether an item is consistent with the rest of the questions that are intended to measure the same thing.
Item–total correlation is the statistical correlation between respondents' answers to one item and their total score on the remaining items in the same scale (usually excluding the item itself; this is called the corrected item–total correlation). It is used when you build a multi-question scale (for example, a 5‑question satisfaction scale) to check whether each question moves in the same direction as the overall construct. High positive correlations mean the item discriminates well and aligns with the scale; low or negative correlations suggest the item may be confusing, irrelevant, or measuring something else.
Usage example
You create a 6‑question service satisfaction scale in Hearo. After collecting 300 responses you run an item analysis and find item 4 has a corrected item–total correlation of 0.10 while the other items are 0.45–0.67. This low value tells you item 4 may not belong in the scale — you might review its wording or translation, or consider removing it.
Practical application
Item–total correlation helps you improve the quality and interpretability of short scales used in surveys: it flags questions that don't fit, which can reduce noise and make scores more reliable. In practice this means clearer results and fewer false conclusions about groups or trends. For Hearo users working with multilingual audiences, item–total correlations are especially useful to detect translation or cultural problems: an item that shows a low correlation in one language group but not others could indicate a wording issue in that translation. Actions after spotting a low correlation include checking translation accuracy, revising phrasing, examining response distributions, or removing the item from the composite score.
FAQ
What is a good item–total correlation value?
Rules of thumb vary, but commonly a corrected item–total correlation above 0.30 is considered acceptable, 0.20–0.30 is weak, and below 0.20 suggests the item may need revision or removal. Use these as guides, not strict cutoffs — consider content importance and sample size.
How does item–total correlation relate to Cronbach’s alpha?
Both assess internal consistency. Items with low or negative item–total correlations often reduce Cronbach’s alpha. You can examine how alpha would change if you removed an item: if alpha rises when an item is removed, that item may be lowering scale reliability.
Does item–total correlation work for yes/no or open-text items?
For binary (yes/no) items use a point‑biserial correlation, which is conceptually similar. Open-text answers need to be coded or translated into categories or numeric scores before you can compute correlations. For qualitative feedback, look for thematic coherence rather than statistical correlations.
Any cautions when using item–total correlations with small or multilingual samples?
Yes. Correlations are less stable with small sample sizes, so interpret low values cautiously. In multilingual surveys, check item–total correlations separately by language: a low correlation in one language may signal a translation or cultural mismatch rather than a bad question overall.