Guide
How do you run a Type 1 gage study (Cg, Cgk) before arguing about parts?
In plain terms, a Type 1 gage study is a short test of the gage on its own: one operator measures one reference part with a known value many times, under the same conditions, and the results give two indices. Cg shows whether the gage's repeatability fits the tolerance. Cgk shows whether repeatability plus bias, the offset between the gage's average reading and the reference value, still fits. Run it before a gage R&R and before any argument about whose readings are right, because R&R does not detect a gage that reads consistently high or low.
By Saola Manufacturing engineering team · Published · Updated
The usual story goes like this. The supplier's report says a bore is in tolerance, receiving inspection rejects some parts at one limit, and both sides send a gage R&R study that passes. Both studies can be correct and the dispute can still be about the gage, because R&R measures scatter, not accuracy. This guide is for the quality engineer or buyer who needs to settle that kind of dispute or prevent it. It covers how to run a Type 1 study, how to calculate and read Cg and Cgk, what a failing result points to, when one reference part is not enough and a linearity check is needed, and what to write on the drawing, RFQ and PO. How to read the R&R results that come after is covered in the gage R&R acceptance criteria guide.
On this page
What is a Type 1 gage study, and what does it catch that gage R&R misses?
It isolates the gage itself by removing the effects of operators and part-to-part variation, so the only things left are repeatability and bias. Gage R&R estimates scatter between readings and between people, but it never compares the readings with a known true value, so a gage that is precise and wrong passes it.
The NIST/SEMATECH e-Handbook defines bias as the difference between a measurement result and its true value, and notes that calibration in an ideal environment does not prove the gage behaves correctly where it is used; temperature, operator, wear and other factors can add bias in the working setup (NIST/SEMATECH e-Handbook, gauge study). A gage R&R uses ordinary production parts whose true values nobody knows, so any constant offset shifts every reading by the same amount and drops out of the variance arithmetic. The study passes, and every reading is still off by the same amount.
Minitab describes the Type 1 study as an evaluation of the variation that comes only from the gage, using one operator and one reference part, and notes that many organizations require it as the first step of a full measurement system analysis (Minitab, what is a type 1 gage study). The convention comes from German automotive practice, and the study is known as a type 1 study, or Verfahren 1, in supplier quality manuals. The usual order is Type 1 first to qualify the gage, then a Type 2 study (a gage R&R with operators) or a Type 3 study (the same without operator influence, for automated gages) to qualify the whole measuring process.
The symptom pattern that points to bias is consistent. Rejections pile up at one tolerance limit and not the other. Two sites disagree by about the same amount on every part. A supplier's capability chart looks centered while yours looks off-center. If you see that pattern, run a Type 1 study before you check the parts again.
How do you run a Type 1 gage study step by step?
Pick one stable reference part with a calibrated value near the middle of the tolerance, have one trained operator measure it repeatedly on the production gage and fixture, re-seating it each time, and record every reading in sequence. Minitab treats 50 readings as an adequate study; fewer readings leave the bias and standard deviation estimates less certain.
- Get a reference part and its value. Use a master, a setting ring, a gage block stack or a real production part whose feature has been measured on a more accurate system, such as a CMM or a calibration lab instrument, with traceable calibration. Write the reference value and its uncertainty on the study sheet; the calibration certificate guide lists what that record should show.
- Check resolution first. Minitab's guideline is a resolution no coarser than 5% of the tolerance (Minitab methods and formulas). A gage that only shows two or three distinct values across fifty readings produces a misleading standard deviation, whatever the result.
- Use the real setup. Measure at the place, with the fixture, probe and program the gage will use in production. A study run on a bench in a temperature-controlled room tells you little about a hand gage used next to the machine.
- Soak the part. Let the reference part and gage reach the same temperature. Dimensions are defined at 20 °C per ISO 1, so state where the study was run and how long the parts soaked.
- Re-seat every reading. Remove and reload the part between readings, and measure at the same marked location so the part's own form error does not enter the result.
- Record in order and plot it. A run chart of the readings against sequence shows drift, steps and outliers that a single standard deviation hides.
- Calculate the mean, standard deviation, bias, a t-test on the bias, Cg and Cgk, as in the next section.
On sample size, Minitab's data guidance asks for 50 measurements for an adequate study (Minitab, data considerations). MoreSteam's description of the same study puts the repetitions at typically fifty or more and draws the run chart limits a typical 10% of the tolerance either side of the reference value (MoreSteam, type 1 gauge study). Readings that wander outside that band are the first sign that the gage, not the part, is the problem.
How are Cg and Cgk calculated, and what limits apply?
Cg compares a fixed share of the tolerance, usually a fifth, with six standard deviations of the repeated readings, and Cgk subtracts the bias from half of that share before dividing by three standard deviations. The common acceptance limit for both is 1.33.
With T the tolerance width, s the standard deviation of the readings, x̄ their mean and xm the reference value, the usual default form is: Cg = (0.2 × T) / (6 × s) and Cgk = (0.1 × T − |x̄ − xm|) / (3 × s). Minitab uses a default of 20% of the tolerance and a six-sigma study variation, both adjustable, and reports the equivalent %Var figures for repeatability and for repeatability plus bias (Minitab methods and formulas). Both settings can be changed, and some customer manuals set their own share of the tolerance, so a result is only comparable when the report states the percentage and the sigma multiplier.
Read the two indices together. A good Cg with a poor Cgk means the gage repeats well but is offset, which can usually be fixed by re-mastering or correcting the zero. A poor Cg means the gage scatters, and no offset correction will help. The bias t-test, a one-sample test of the null hypothesis that the bias is zero, tells you whether the offset is real or just noise from a small sample. The international route to the same decision is ISO 22514-7:2026 (which replaced the 2021 edition in September 2026), which combines the uncertainty components following the GUM approach and expresses the result as a capability figure (CMS, CMP) or a capability ratio (QMS, QMP) rather than as Cg and Cgk.
| Item | Typical requirement | What a miss means |
|---|---|---|
| Readings on one reference part | 50 (Minitab data guidance) | Standard deviation and bias estimate too uncertain to judge |
| Gage resolution | 5 % of tolerance or finer | Readings collapse into a few values; s is unreliable |
| Share of tolerance (K) | 20 % (Minitab default; customer manuals may differ) | State it; results with different K are not comparable |
| Cg | 1.33 or more | Repeatability too large for the tolerance |
| Cgk | 1.33 or more | Repeatability plus bias too large; check zero and master first |
| Bias t-test | p-value 0.05 or more at the conventional 5 % significance level (bias not significant) | The offset is real; correct it or account for it |
| %Var repeatability (K / Cg) | 15 % or less at K = 20 % | Same message as Cg, in percent form |
The worked example below is illustrative, not test data: one bore gage on a typical 25 mm bore with a typical total tolerance band of 0.050 mm. The gage passes Cg and a typical tolerance-based R&R check, and fails Cgk because of a bias of 0.002 mm. That is the situation in which both sides of a dispute hold a passing R&R study.
| Quantity | Value | Reading |
|---|---|---|
| Reference value of the master ring | 25.010 mm | Input, from calibration |
| Mean of 50 readings | 25.012 mm | Input |
| Standard deviation s | 0.0010 mm | Input |
| Bias | +0.002 mm | t = 14.1, clearly significant |
| Cg = 0.2 × 0.050 / (6 × 0.0010) | 1.67 | Pass |
| Cgk = (0.005 − 0.002) / (3 × 0.0010) | 1.00 | Fail |
| Repeatability share of tolerance, 6s / T | 12 % | Would look acceptable in an R&R on tolerance basis |
| Smallest tolerance this gage supports at Cgk 1.33 (40s + 10 × |bias|) | 0.060 mm | Wider than the drawing’s 0.050 mm |
| Smallest tolerance after the bias is corrected (40s) | 0.040 mm | Fits once re-mastered |
In practice the high-reading gage rejects good bores just under the upper limit and accepts undersize bores just below the lower limit. That pattern matches a dispute in which one side rejects at one limit and the other side does not. Rearranging the Cgk formula gives the smallest tolerance a gage can support, which is a quick way to answer whether a gage is good enough for a drawing before anyone measures a production part.
What makes a gage fail Cg or Cgk, and how do you fix it?
A failed Cgk with a good Cg is almost always a setting problem: a wrong zero, a worn or mis-valued master, or a temperature offset. A failed Cg is a hardware or method problem: resolution, fixturing, contact force, or a gage that is simply too coarse for the tolerance.
Work through the causes in this order, because the first ones are the cheapest to check and the most common. Look at the run chart before changing anything, because the pattern of the readings usually points to the cause.
| What the study shows | Most likely cause | Fix to try first |
|---|---|---|
| Cg good, Cgk poor, readings steady | Zero or master offset; wrong reference value entered | Re-master, check the master’s certificate value, rerun |
| Bias reverses between two masters of different size | Linearity error across the range | Run a linearity check; restrict the gage to the range it passes |
| Readings trend up or down with sequence | Warming of part or gage by handling; drift | Soak longer, use gloves or a stand, rerun with breaks |
| Readings sit on only a few values | Resolution too coarse | Use a finer-reading gage; do not trust s from the study |
| Cg poor, random scatter | Part not seated the same way; contact force; dirt | Improve the fixture or locating stop; clean; constant-force gage |
| One or two wild readings | Mis-seat, burr, chip on the anvil | Find the reason before deleting any reading; then rerun |
| Cg poor even with good method | Gage class too coarse for the tolerance | Change gage type, or ask whether the tolerance is needed |
Two warnings apply. First, do not drop readings to make Cg pass; an outlier is a fact about the method, and deleting it hides the problem the study was meant to find. Second, a gage that cannot pass at the drawing tolerance is not always the gage's fault. If the smallest supportable tolerance is close to the drawing tolerance with the best practical gage, the tolerance itself may be the real problem, a cost question covered in the 10:1 rule and measurement uncertainty guide.
When is one reference part not enough, and how do you check linearity?
A Type 1 study proves the gage at one size only, so a gage that measures several part sizes, or a wide tolerance band, also needs a linearity check. Linearity shows whether the bias changes across the operating range.
The AIAG Measurement Systems Analysis manual treats bias and linearity as separate studies from R&R. In a linearity study, reference parts with known values are chosen across the range the gage is used for, each is measured repeatedly, the bias at each reference value is calculated, and bias is regressed against reference value. A slope significantly different from zero means the gage reads differently at different sizes. A typical layout is about five reference parts measured ten or more times each, in random order so the operator cannot remember earlier readings.
Where linearity matters most in machined parts:
- One adjustable gage, several part numbers. A bore gage or height gage set on one master and used across a family of sizes.
- Long-range instruments. Calipers, height gages and vision systems used near the ends of their travel or field of view.
- Critical limit away from the reference. If the dispute is about one limit, put the reference part near that limit, or run Type 1 at both ends.
A simple alternative that suppliers handle well is a Type 1 study at each size the gage will actually measure, with the reference part for each one stated. It takes more readings, but each result is easy to read and to repeat later.
What should the drawing, RFQ and PO say about a Type 1 study?
List the characteristics that need it, the criteria and their basis, the reference part, and the raw readings you expect back. A note that only says “MSA required” leaves the method, the criteria and the scope to the supplier.
- Scope. “Type 1 gage study required for characteristics flagged ◆ on the drawing” (critical fits, sealing bores, datum features), not for every dimension.
- Criteria and basis. “Cg and Cgk at least 1.33, K = 20% of tolerance (the typical default), six-sigma study variation, bias t-test reported,” or cite your customer manual, so the result cannot be recalculated on a friendlier basis.
- Reference part. Who provides it, how its value is established and with what calibration, and where in the tolerance it sits.
- Setup. The gage, fixture and location used for production acceptance, and the soak condition.
- Deliverables. All readings in sequence, the run chart, and the calculation, not only the summary. Then the gage R&R, which is read using the criteria in the gage R&R guide.
- Referee rule. Which measurement method settles a borderline part if supplier and customer readings disagree after both gages pass.
If a dispute is already open, the order is the same: Type 1 on each side's gage using a shared, calibrated reference part; then correlation on the same serialized parts; then decide which readings govern. The first step often settles it, because a bias shows up at once and is simple to correct. For the correlation step, measure the same serialized parts (ten or more, spread across the tolerance) on both gages at the same marked locations and soak condition, then look at the difference for each part: a roughly constant difference is an offset between the gages, a difference that grows with size is a linearity problem, and a difference that jumps about is a repeatability or fixturing problem on one side. Write down which gage and which reference part govern before the next lot ships.
How is gage capability handled on a Saola order?
Machined features are made and inspected at our lead CNC plant, and a Type 1 study on a named characteristic can be requested on the RFQ and quoted. For cast, stamped or finished parts, the casting, stamping and finishing are done by partner plants we manage, and the quote names the production site and the country of origin.
Saola Manufacturing is the brand of Aximora Global Trading Co., Ltd., a managed manufacturing partner based in Ho Chi Minh City, not a factory. Turning and milling run at our lead CNC plant, T.K Viet Nam Precision Company Limited, as described on the CNC milling and CNC turning pages. Its inspection equipment includes a 2D vision measuring machine, a height gauge and a hardness tester; the quality page explains how lots are checked. If your drawing needs a Type 1 study, say on the RFQ which characteristics, which gage type you expect and which criteria apply, and the quote will say how it would be done and with which reference. Features formed by die casting or changed by finishing are measured after that step, so state on the RFQ whether the study applies before or after finishing. ISO 9001 certification is in progress, not held. Send the drawing and your gage study requirement through the quote form.
Frequently asked
Is a Type 1 gage study the same as a gage R&R?
No. A Type 1 study checks the gage alone, with one operator and one reference part of known value, and reports repeatability and bias. A gage R&R uses several operators and production parts and reports repeatability and reproducibility, but not bias.
Why is 1.33 the usual limit for Cg and Cgk?
It mirrors the familiar process capability target of 1.33 and leaves margin for the variation a Type 1 study does not include, such as operators and part form. It is a convention, so check whether your customer manual sets a different value.
Can I run a Type 1 study on a production part instead of a master?
Yes, if the feature has been measured on a more accurate, calibrated system to give it a reference value, and the measured location is marked. Its form error then stays out of the result.
What if Cg passes but Cgk fails?
The gage repeats well but reads off-center. Check the zero, the master and its certificate value, and the temperature, correct the offset, and rerun the study before anyone measures parts again.
Does a Type 1 study apply to go/no-go gages?
No. Cg and Cgk need variable readings. Fixed-limit gages are controlled by their own gage tolerances and calibration and by attribute studies, as covered in the go/no-go thread gauge guide.
Sources
- Methods and formulas for Type 1 Gage Study, Minitab, accessed October 2026
- What is a type 1 gage study?, Minitab, accessed October 2026
- Type 1 Gauge Study, MoreSteam, accessed October 2026
- NIST/SEMATECH e-Handbook of Statistical Methods, section 2.4.5, Analysis of bias, NIST, 2012
- ISO 22514-7:2026, Statistical methods in process management, Capability and performance, Part 7: Capability of measurement processes (catalog record with scope and edition history), EVS, accessed October 2026
- Data considerations for Type 1 Gage Study, Minitab, accessed October 2026
- Measurement Systems Analysis (MSA), 4th edition, Automotive Industry Action Group, 2010