Content Validity Index (CVI) Calculator
Paste your expert panel's relevance ratings — 1 to 4 for every item — and get each item's I-CVI and chance-adjusted modified kappa, the scale-level S-CVI/Ave and S-CVI/UA, and whether each item meets Lynn's (1986) minimum for a panel of your size, with the working shown for any item you pick. Free, no sign-up, and the ratings never leave your browser.
Paste from a spreadsheet or type ratings separated by commas or spaces: 1 = not relevant, 2 = somewhat relevant, 3 = quite relevant, 4 = highly relevant. Blank cells are skipped.
Results
Paste the panel's ratings, or load the example, to see every item's I-CVI and the scale-level CVIs.
How to Use the CVI Calculator
- 1. Collect the ratings. Ask each expert to rate every item's relevance to the construct on the 4-point scale Davis (1992) labeled: 1 = not relevant, 2 = somewhat relevant, 3 = quite relevant, 4 = highly relevant.
- 2. Paste them in. The default layout is one row per expert and one column per item. If your sheet has one row per item, choose One item under Each row of the table is. Names in the first row and first column are detected for you.
- 3. Read the item table. Each item shows how many experts rated it 3 or 4, its I-CVI, the chance agreement Pc, the modified kappa k* with its evaluation, and whether it meets Lynn's minimum. Pick any item to see its working.
- 4. Report the scale. Lead with S-CVI/Ave, give S-CVI/UA beside it, report the range of the I-CVIs, and download the item table as CSV for an appendix.
How the Content Validity Index Is Calculated
The ratings are dichotomized: 3 and 4 count as relevant, 1 and 2 as not relevant. Everything else follows from counting.
- I-CVI = A ÷ N, where A is the number of experts who rated the item 3 or 4 and N is the number who rated it (Lynn, 1986). Four of five experts gives 0.80.
- S-CVI/Ave = the mean of the I-CVIs (Polit & Beck, 2006). With no blank ratings it equals the average share of items each expert called relevant, which the tool lists per expert.
- S-CVI/UA = the share of items that every expert rated 3 or 4 — universal agreement.
- Pc = N! ÷ (A! × (N − A)!) × 0.5ᴺ: the probability that exactly A of N experts call the item relevant if each decides by tossing a fair coin.
- k* = (I-CVI − Pc) ÷ (1 − Pc), the modified kappa of Polit, Beck and Owen (2007), which discounts the agreement a small panel could reach by chance.
Worked Example: Six Experts Rate Eight Items
Press Load example. Six experts rated eight draft items. Four items were rated 3 or 4 by all six, two by five, one by four and one by three:
| Items | Rated 3 or 4 | I-CVI | Pc | k* | Evaluation | Lynn (6 experts) |
|---|---|---|---|---|---|---|
| 1, 2, 3, 5 | 6 of 6 | 1.00 | 0.016 | 1.00 | Excellent | Meets |
| 6, 7 | 5 of 6 | 0.83 | 0.094 | 0.82 | Excellent | Meets |
| 4 | 4 of 6 | 0.67 | 0.234 | 0.56 | Fair | Below |
| 8 | 3 of 6 | 0.50 | 0.313 | 0.27 | Poor | Below |
The working panel opens on item 4, the first item without universal agreement: Pc = 6! ÷ (4! × 2!) × 0.5⁶ = 15 × 0.015625 = 0.234, and k* = (0.667 − 0.234) ÷ (1 − 0.234) = 0.56. At the scale level, S-CVI/Ave = (4 × 1.00 + 2 × 0.83 + 0.67 + 0.50) ÷ 8 = 0.85 and S-CVI/UA = 4 ÷ 8 = 0.50 — both short of the usual targets.
Items 4 and 8 are the ones to revise or drop. Without them the remaining six items reach S-CVI/Ave = 0.94 and S-CVI/UA = 0.67, the pattern Polit and Beck (2006) warned about: the averaging index clears 0.90 once the weak items go, while universal agreement stays hard to reach with six experts, because one dissenting rating removes an item from it.
What Counts as Good Content Validity
| Index | Common benchmark | Source |
|---|---|---|
| I-CVI, 3–5 experts | 1.00 — every expert rates the item 3 or 4 | Lynn (1986) |
| I-CVI, 6–10 experts | 5 of 6, 6 of 7, 7 of 8, 7 of 9, 8 of 10 — never below 0.78 | Lynn (1986); Polit & Beck (2006) |
| k* | Excellent above 0.74, good 0.60–0.74, fair 0.40–0.59 | Polit, Beck & Owen (2007) |
| S-CVI/Ave | 0.90 or higher | Polit & Beck (2006), after Waltz, Strickland & Lenz (2005) |
| S-CVI/UA | 0.80 or higher, a strict standard | Davis (1992), as read by Polit & Beck (2006) |
The two item-level rules can disagree. With five experts, Lynn requires all five to agree, but four of five gives k* = 0.76, which Polit, Beck and Owen rate excellent — and they concluded that an I-CVI of 0.78 or higher with three or more experts is evidence of good content validity. The calculator reports both, so state which rule you applied. Beyond ten experts, Lynn's table ends and the tool applies the 0.78 floor.
A CVI measures agreement about relevance, nothing more. It says nothing about whether the items are clear, whether the set covers the whole construct, or how the scale behaves with real respondents — ask experts about clarity and coverage too, and follow content validation with pilot data and a reliability analysis.
Frequently Asked Questions
- What is a good content validity index?
- For items, Lynn (1986) required an I-CVI of 1.00 with three to five experts and set minimums from 5 of 6 to 8 of 10 for larger panels, never below 0.78. For the whole scale, Polit and Beck (2006) recommend an S-CVI/Ave of 0.90 or higher, and 0.80 is a common, strict target for S-CVI/UA.
- How many experts do I need for a content validity index?
- Lynn advised at least three and suggested that more than ten is probably unnecessary. Many studies use five to ten. More experts make each item's I-CVI less sensitive to one unusual rating, but they also make universal agreement, and so S-CVI/UA, harder to reach.
- What is the difference between S-CVI/Ave and S-CVI/UA?
- S-CVI/Ave averages the item I-CVIs, so an item rated relevant by five of six experts still contributes 0.83. S-CVI/UA only counts items that every expert rated 3 or 4, so it falls as the panel grows. Polit and Beck (2006) prefer S-CVI/Ave and recommend reporting which method you used.
- Why calculate a modified kappa as well as the I-CVI?
- With few experts, high agreement can happen by chance: two experts both say relevant a quarter of the time even if they guess. Modified kappa subtracts the binomial probability of the observed agreement, so it shows whether an I-CVI is impressive for a panel of that size.
- What is the difference between CVI and CVR?
- The content validity ratio (Lawshe, 1975) asks experts whether each item is essential, useful but not essential, or not necessary, and computes CVR = (nₑ − N/2) ÷ (N/2), where nₑ is the number who chose essential. The CVI asks about relevance on a 4-point scale instead. They answer related questions and are not interchangeable; this calculator computes the CVI.
- Can I use a 3-point or 5-point relevance scale?
- Lynn recommended four points precisely to avoid an ambivalent middle option, and the I-CVI only distinguishes relevant from not relevant. If your panel used another scale, decide which ratings mean relevant, recode them to 4 and the rest to 1, then paste the recoded table.