Quick Navigation
Let's cut the fluff: precision and accuracy in laboratory analysis are not synonyms. I've spent a decade working in analytical labs, and the amount of confusion around these two terms is unreal. That's why I'm writing this guide—to give you a practical, no-nonsense breakdown of what they mean, how to evaluate them, and how to improve both in your own lab.
What's the Real Difference Between Precision and Accuracy?
Think of shooting at a target. Precision is whether your shots cluster together. Accuracy is whether they hit the bullseye. You can be precise (tight cluster) but miss the center entirely—that's bias. You can be accurate on average, but with shots scattered everywhere—that's high variability.
In lab terms, precision measures how reproducible your results are. Run the same sample 10 times, and precision tells you how close those 10 results are to each other. Accuracy tells you how close they are to the true value. A method can be precise but not accurate, accurate but not precise, both, or neither.
The statistical definitions are straightforward. Precision is usually expressed as standard deviation, variance, or coefficient of variation. Accuracy is expressed as bias, which is the difference between the measured mean and the true value. In international standards like ISO 5725, they use 'trueness' and 'precision', where trueness is the closeness of the mean of measurements to the true value. That terminology can confuse people—but it's useful to know.
| Aspect | Precision | Accuracy |
|---|---|---|
| Definition | Closeness of repeated measurements to each other | Closeness of a measurement to the true value |
| Key metric | Standard deviation (SD), RSD | Bias, percent recovery |
| Improving | Reduce random errors | Reduce systematic errors |
| Example | Same reading every time, but wrong | Scattered readings, but centered on true value |
I've seen too many analysts chase a ridiculously low RSD and call it day, not realizing their method is consistently biased. That's like polishing a toilet seat—doesn't change what's underneath.
Why Both Matter in Laboratory Analysis
If you're working in clinical diagnostics, environmental monitoring, or pharmaceutical QC, having both precision and accuracy is non-negotiable. A test that's precise but inaccurate can lead to misdiagnosis. Think about glucose meters: if a meter is consistently 10% off, it might be precise (same reading each time) but inaccurate (the reading is wrong). For a patient, that's a disaster.
Early in my career, I worked in a hospital lab. Our glucose analyzer had a +5% bias. It was precise—same reading every time—but it read 5% high. That meant patients' real glucose was 5% lower than what doctors saw. We caught it only after a diabetic patient slipped into hypoglycemia while our QC looked fine. That's the danger of precision without accuracy.
Imagine you're testing water for lead. A precise but inaccurate method might consistently report 10 ppb when the true level is 8 ppb. That could trigger a false violation. The legal and financial consequences are real.
How to Evaluate Precision and Accuracy in Your Lab
Here's how I personally evaluate a method before using it for reporting. I'll give you the exact steps and formulas I use.
Calculating Standard Deviation and RSD
Run at least 6 replicates of a real sample. Calculate the mean, standard deviation (SD), and relative standard deviation (RSD). RSD = (SD/mean) × 100%. A RSD below 2% is typically excellent, but you should compare against method requirements. For trace analysis, I'm happy with
I once worked with a method that had an RSD of 0.8%, which sounds great until I realized the concentrations were near the detection limit—the precision was actually noise. So always check the concentration before celebrating.
Measuring Bias and Recovery
Use a certified reference material (CRM). Measure it 3-5 times and calculate the average. Percent recovery = (measured/certified) × 100%. If recovery isn't within 100±10%, you have accuracy issues. Also spike real samples to check matrix effects—that's often where accuracy gets murdered.
Control charts are your best friend. Plot QC data daily on an X-bar chart. When creating control charts, I recommend using a Levey-Jennings chart for QC data. Plot each QC measurement, draw the mean and ±2SD, and look for trends. A run of 5 points on one side of the mean is a sign of drift. That's your early warning system. I once caught a gradual drift in a pH meter before any result went out of spec. For more details, refer to ISO 5725-1 and CLSI EP05-A2.
Common Mistakes That Ruin Precision and Accuracy
I've seen these mistakes destroy precision and accuracy in every lab I've worked in. Here are the top 5 killers.
- Ignoring environmental conditions – temperature swings cause expansion of glassware and affect sample volumes. One summer, our AC broke, and results became erratic until we fixed it.
- Using uncalibrated pipettes – a tiny misdelivery at low volumes can be huge. Calibrate them routinely.
- Contamination during sample prep – you think you're precise, but you're just consistently contaminating.
- Not mixing samples properly – if your sample is inhomogeneous, no amount of analytical skill will save you.
- Incorrect statistical treatment – some people just report the mean and ignore outliers. That's wrong.
One subtle issue is the use of plastic pipettes for volatile solvents—your precision suffers because the volume changes with evaporation. Also, don't forget to condition your pipette by drawing and dispensing the liquid a few times before measuring. It's a tiny thing that can make a huge difference.
And don't get me started on labs that never run blanks. That's not precision; that's self-deception.
Practical Steps to Improve Precision and Accuracy
Here's a step-by-step system I've built over the years. Implement these and you'll see immediate improvements.
| Step | Action | Why It Helps |
|---|---|---|
| 1 | Calibrate instruments and pipettes on a strict schedule | Eliminates most systematic errors |
| 2 | Run matrix-matched QC samples with every batch | Detects bias and drift early |
| 3 | Write detailed SOPs and follow them exactly | Reduces human error and variability |
| 4 | Train and retrain your staff | Humans are the biggest source of variation |
| 5 | Monitor control charts and take corrective action | Prevents problems before they affect results |
Your calibration schedule should be based on manufacturer recommendations and your own data. If you see a drift, increase the frequency. I always write down calibration dates on the instrument itself, so there's no excuse.
One piece of advice: don't just buy expensive instruments and think you're done. A $200k GC-MS will give you garbage if you don't handle your samples right. The instrument is only part of the equation.
Frequently Asked Questions
How do I fix a consistent 5% bias in my ICP-MS results?
Start by verifying your calibration curve. Are your standards fresh? Are you using matrix-matched standards? Bias often comes from calibration errors or internal standard recoveries. Check your CRM too—if it's out of date, that's a clue. I once traced a bias to a contaminated internal standard. Simple replacement fixed it.
What is the maximum acceptable RSD for a pesticide residue method by GC?
There's no universal answer—it depends on the concentration and the method's intended use. For regulatory work, I typically aim for RSD
Why is my precision excellent but my accuracy poor in ELISA?
This happens when you have a systematic error, often from the standard curve or the dilution factor. If your precision is tight but the results are off, suspect your standards, pipetting of standards, or the plate reader's wavelength. Also, check for matrix interference in your samples. I've seen ELISA bias caused by a bad fit of the standard curve with a linear model when a four-parameter logistic fit was needed.
This guide is based on my hands-on lab experience and aligns with ISO 5725-1 and CLSI EP05-A2 guidelines.
