Explain the T- Test?
- What is a p-value?
- What is a t-value?
Is a statistical test used to compare the mean of two groups of data.
P-value is the probability value - the likelihood of finding a mean difference if there is no difference in the population.
If p-value is less than 0.05 it means that the test hypothesis is false and data is statistically different
T-value tells you how far apart the two means are
What are the statistical rules for control charts ?
Nelson’s Rules
R1 single point outside control limits
R2 8 consecutive points either side of the mean
R3 2 out of 3 results above or below control limits
R4 4 out of 5 results above or below control limits
R5 6 results in a row increasing or decreasing
What range would you expect for an assay method?
95-105% of test concentration
What value would you expect for LOD?
Signal to noise ratio of 3:1
In what circumstances would you use the following statistical methods:
T test
2 sided t-test
F-test / ANOVA
T-test 1 one set of data against a specification
2 sided t-test- comparing two sets of data
F-test / ANOVA- multi-variate analysis- used to assess multiple factors
What statistics are behind sampling tables as per ISO2859?
What is consumer risk vs cooperate risk?
Sampling tables use lot size, inspection levels and AQL’s to determine a maximum allowable defects in a particular batch
Consumer risk is risk of accepting a batch of unsatisfactory product vs cooperate risk of rejecting a batch of satisfactory quality.
Give examples of different types of sampling
Simple Random - all equal chance
Systematic Sampling - start / middle/ end
Stratified sampling- put in groups, sample in each group
What rules do we have for extrapolation of data?
Minimum of 3 batches
Variability of data needs to be understood - no significant changes at accelerated conditions
Can’t extrapolate more than 12 months beyond last data point
Or 6 moths for refrigerated documents
Define the following and when they are used?
CP
CPK
PP
PPK
PROCESS C Capability (VALIDATION )
CP - evaluates spread of data compared to specification limits- assuming data is centered
CPK- evaluates spread of data takes in to account process mean and it’s proximity to specification
PROCESS PERFORMANCE (CPV)
Pp - same as above but used more data
Ppk- same as above but used more data
What is the key value for CP CPK PP PPK?
> 1.33
If CP / PP Are the same short / long term variation is the same
When applying process capability to control charts what calculation do you use?
CP / PP - upper limit - lower limit / 6 x Standard Deviations
CPK / PPK - (UCL - MEAN/ 3 SD ) X (MEAN - LCL / 3SD)
How do you calculate Accuracy?
Recovered Vale / Theoretical x 100 = % agreement with true value
How to Calculate Precision:
Repeatability
Intermediate Precision
Reproducibility
Repeatability- 3 times at 3 levels or 6 times at 100% - calculate relative Standard Deviation = SD / Mean x 100 - typically 2% or less is acceptable
Intermediate precision- different laboratories, analysts, days etc and calculate RSD above
What is the regression value for linearity?
0.996 minimum
What is a shewhart chart vs a CUSUM chart vs a Pareto chart?
Shewhart is standard control chat - upper and lower action and alert limits - each point is a single measurement
CUSUM - detects small gradual shifts by summing up deviations from target value
Pareto - break down into groups - bar chart in ascending/ descending order
How do you use ISO2859?
Understand product- critical, major and minor
Determine batch size and select inspection levels (normal reduced tightened)
Determine AQL - typical value is 0 - critical, 2.5- major and 4.0 - minor
Use table to determine accept / reject
How do you apply 95% confidence intervals to control charts?
UCL 95% - mean + 1.96 x SD
LCL 95% - mean - 1.96 x SD
1.96 is Z value for 95% confidence for normal distribution
