Understanding and Interpreting Statistics in Assessments Clare Trott and Hilary Maddocks This Session • Why use statistics in assessments? • “Averages”, Standard Deviation, variance, Standard Error • Normal distribution, confidence intervals • Scales • Overlapping confidence intervals • Why use statistics in assessments? • What are the assumptions made? Feedback MEDIAN MEAN MODE Central Tendency AVERAGE Spread Which is better? When? What is • Standard Deviation? • Variance? • Standard Error? Feedback MODE MOST OFTEN D E MEDIAN MED IAN Means Make Everyone Add Numbers (and) Share Standard Deviation 2, 3, 6, 9, 10 Mean = 6, SD = 3.16 2, 2, 6, 10, 10 Mean = 6, SD = 3.58 • Measures the average amount by which all the data values deviate from the mean • Measured in the same units as the data Variance and Standard Deviation Variance = 𝜎 2 Σ 𝑥−𝑥 = 𝑛 2 Mean (𝑥 ) Standard deviation = σ= Σ(𝑥−𝑥)2 𝑛 Standard Error This is the variance per person 𝜎2 𝑆𝐸 = 𝑛 Normal Distribution Confidence Intervals • What is Normal Distribution? • What are Confidence Intervals? • Why is it useful? • Why are they important? Feedback Normal Distribution Number of standard deviations from the mean Confidence Intervals TRUE SCORE 99% 95% Confidence Confidence Interval Interval CI The wider the range the more confident we can be that the true score lies in this range Due to inherent error in measurement it is better to quote a 95% confidence interval Confidence Intervals 1.645 -1.645 -1.96 1.96 -2.575 2.575 90% Confidence Interval 95% Confidence Interval 99% Confidence Interval 95% Confidence Intervals True Score • True score lies inside CI 95% of occasions • 1 in 20 (5%) will not include the true score Scales • What scales are used in reporting? • How are they defined? • Why are standardised scores preferred? Feedback Scaled scores 4 6 8 10 12 14 16 Percentiles 2 10 25 50 75 90 98 120 130 Standardised scores 70 Very low 80 low 90 Low average 100 average 110 High average high Very high Simplified Table standar dised percent ile scaled 130 and above 98th >16 + 3SD Within top 2% Very high 120-129 91-97 14-15 + 2SD Above 91% high 110-119 75-90 12-13 + 1SD Above 75% High average 90-109 25-74 8-11 Mean Above 25% average 80-89 10-24 6-7 -1SD Above 16% Low average 70-79 2-9 4-5 -2SD Above 10% Below average Below 70 Below 2 <4 -3SD Lowest 2% Very low Scale to Standardised • • • • 1 to 5 ratio 10 scaled 9 scaled 11 scaled • 15 scaled • 6 scaled 100 standardised 95 standardised 105 standardised 125 standardised 80 standardised Standardised scores against standard deviations mean -1sd 1sd -2sd 2sd 3sd -3sd 70 Very low 80 low 90 Low average 100 average 110 High average 120 high 130 Very high Percentiles against standard deviations mean -1sd 1sd -2sd 2sd -3sd 3sd 2 Very low 10 low 25 Low average 50 average 75 High average 90 high 98 Very high Scaled scores against standard deviations mean -1sd 1sd -2sd 2sd -3sd 3sd 4 Very low 6 low 8 Low average 10 average 12 High average 14 high 16 Very high Differences in Class Intervals Suppose we have the class intervals for two tests which could be linked, and we wish to find whether there is a significant difference between the two sets. Test 1 95% Confidence Interval 102 ± 15.8, standard error 2.96 Test 2 95% Confidence Interval 118 ± 23, standard error 6.63 86.2 102 105 117.8 118 131 There appears to be no significant difference as there is a distinct overlap. H0 : There is no significant difference in the two Confidence Intervals (the new confidence interval contains zero) H1 : There is a significant difference in the two Confidence Intervals (the new CI does not contain zero Formula Difference in scores ±1.96 𝑆𝐸12 + 𝑆𝐸22 =(118 – 102) ±1.96 2.962 + 6.632 = 16 ± 14 New CI 2 16 30 This does not contain zero so we reject H0 and so there is a significant difference in the two tests. Test 1 95% Confidence Interval 95 ± 6, standard error 3.06 88 95 102 Test 2 95% Confidence Interval 106 ± 10, standard error 5.102 96 106 116 There appears to be no significant difference as there is a distinct overlap. H0 : There is no significant difference in the two Confidence Intervals (the new confidence interval contains zero) H1 : There is a significant difference in the two Confidence Intervals (the new CI does not contain zero Difference in scores =(106 – 95) ±1.96 𝑆𝐸12 + 𝑆𝐸22 ±1.96 3.062 + 5.1022 = 11 ± 11.6 New CI -0.6 11 22.6 This does contain zero so we accept H0 and so there is no significant difference in the two tests.