How many points are appropriate for creating a standard curve?
Thread Content
How many points are needed for the calibration curve? The number of data points required for a standard curve depends on the concentration range of the component being analyzed, the response characteristics of the analyzer, interfering factors, and the relationship between concentration and the type of detection signal response. 1. 3 points: For some low concentrations, especially in trace analysis where the concentration range is not very wide, and the detector response is reliable with minimal background interference, it is sufficient to use a smaller number of working points; generally, 3 concentration points are enough, and in some cases even just one concentration point is sufficient, with the origin being used as another point. 2. 4–6 data points: When the range of sample concentrations measured is wide, and the detector response does not follow a perfect linear curve, a quadratic curve or segmented correction methods can be used to reduce data deviations. In such cases, more data points are needed; however, if segmentation is not used, 4–6 data points are sufficient. 3. At least two points: but in the case of segmented correction, each segment requires at least two data points. For some samples that do not follow a specific concentration range pattern, especially in certain testing laboratories, if the analyzer’s detection response is reliable and there is little influence from environmental factors, it is possible to create a calibration curve. If environmental factors have a significant impact, there is no need to create a calibration curve; instead, the standard addition method or the standard addition-with-one-dilution method is simpler. In fact, we usually use five points (excluding zero concentration) ； Generally, the first point is set at 5 to 10 times the detection limit, and thereafter the values are increased by 1 time each; it is appropriate for the highest concentration to be 10 to 20 times the lowest concentration (taking physical and chemical analysis experiments as an example). Of course, it needs to be adjusted according to the sensitivity of the instrument. Refer to GB/T22554-2010 \"Linear calibration based on standard samples\": (1) The concentration range of the standard curve should cover the range of the quantity being measured under normal operating conditions ； (2) The components of the standard sample should be as consistent as possible with those of the sample being tested ； (3) The concentration values of the standard samples should be distributed at equal intervals within the measurement range ； (4) The number of standard samples should be at least 3 concentrations ； (5) Each standard point should be repeated at least 2 times, and this repetition refers to the time starting from dilution ； If the **standard recommends a specific concentration range, try to follow** that standard. In our work, we often use linear calibration because linear equations are the simplest.
The key is to consider the required detection accuracy; at least three points, right? However, many small and medium-sized enterprises use a two-point comparison method, considering that a absorbance value lower than that of the standard sample means the test is passed; this might be acceptable for process control, but it seems problematic as a result for product testing.
The principle of linear fitting is to determine what level of linear correlation the person using this so-called standard curve requires. For some exploratory or conceptual data analyses, a correlation coefficient of 0.8 is already considered quite good. For methods with well-established standard analytical procedures, when conducting routine analysis, it is appropriate to require a linear correlation coefficient of 0.9 or higher. For some analysis methods that perform well both in theory and in practice, the base linear correlation can reach 0.99 or even higher. From the perspective of linear correlation, in the first case, the more data there is, the better; from a statistical standpoint, it is ideal to have 20 sets of data. In the second case, it is necessary to take into account the flash deviation of linear data that meets statistical standards; for a 95% confidence level, at least 5 sets of data are required. In the third case, the methods are already reliable and mature; for skilled operators, most analysis methods may require only three sets of data, and these data actually serve as a verification of the standard curve. For some specific and reliable analysis methods, such as Coulomb analysis, one data point may be sufficient, as Faraday’s law of electrolysis serves as its theoretical basis. Furthermore, as mentioned by the colleague upstairs, the nature of the analysis to be carried out also plays a role in this decision. Quality inspections of products entering and leaving the factory, as well as routine monitoring and analysis of processing equipment, require different levels of precision in terms of data; consequently, the linear relationship of the standard curves will also differ. As for which set of data points to use, you may need to decide based on the actual circumstances.