 Home
 Studying in New Zealand
 Qualifications and standards
 NCEA
 Māori & Pasifika

Providers and partners
 About education organisations
 NZQA's quality assurance system for tertiary education organisations
 Quick links to NZQF documents
 Approval, accreditation and registration
 Consistency of graduate outcomes
 Selfassessment
 External evaluation and review
 Assessment and moderation
 Development of assessment standards
 Submitting results and awarding qualifications
 The Education (Pastoral Care of International Students) Code of Practice
 Offshore use of qualifications and programmes
 Guidelines and forms
 About us
Assessment Report
Level 3 Statistics 2016
Standards 91584 91585 91586
Part A: Commentary
Candidates who performed well across the standards were able to demonstrate understanding of statistical and probability concepts, integrating these understandings with contextual knowledge when required. Successful candidates demonstrated both calculationbased and analyticalbased approaches.
When using contextual knowledge as part of a response, candidates needed to link this knowledge to an aspect of statistical knowledge. Contextual knowledge or speculation alone was not sufficient.
Part B: Report on Standards
91584: Evaluate statistically based reports
Candidates who were awarded Achievement commonly:
 understood the principles of experimental design, e.g. use of random allocation to assign treatments to experimental units
 constructed confidence intervals from single survey percentages
 understood margin of error, e.g. explaining the necessity for a margin of error to be included in statistical reports, calculating the margin of error using the ‘rule of thumb’, appreciating the relationship between sample size and the size of the margin of error
 described features of a statistical study design, e.g. explanatory and response variables, difference between observational study and an experiment.
Candidates who were assessed as Not Achieved commonly:
 gave vague responses that were not in context or not linked to statistical concepts
 confused statistical terms, e.g. sample with population, random allocation with random selection, sample error with nonsampling error
 were unable to construct a confidence interval
 wrote responses that were not guided by the question asked, e.g. discussing “using the past as a source of data” when asked to discuss “extending the results inappropriately”
 confused features of experiments and observational studies
 believed nonsampling errors or sources of bias could be removed completely, e.g. random allocation “removes bias” rather than minimising bias
 made incorrect inferences about the survey respondents rather than the target population
 could not explain the need for margin of error in a statistical report and did not understand the relationship between sample size and margin of error.
Candidates who were awarded Achievement with Merit commonly:
 explained how random allocation reduces bias or creates two fair or balanced groups, using specific contextual examples of bias
 identified confounding variables for observational studies and illustrated with contextual examples drawn from the text
 interpreted correctly a confidence interval in context to make an inference about a (target) population
 recognised the difference between a proportion versus count in a report
 used the relationship between sample size and the margin of error to reason correctly about two different studies based on their reported margin of errors
 linked survey table results with report statistics
 used understanding of study design to support comments on the nature of the claims made.
Candidates who were awarded Achievement with Excellence commonly:
 responded to a claim concerning the difference between two percentages by constructing and interpreting a confidence interval, and explaining that a claim could not be made because 0 was included in the interval, or because both values of the confidence interval were either both positive or both negative
 described potential issues with the survey design and described how the issues could affect a specific aspect of the report (e.g. conclusions or claims presented) in context with examples to illustrate the issue
 understood the ‘rule of thumb’ acceptable percentage range (30% to 70%) and implications for margin of errors of survey percentages outside of this range (when constructing approximate 95% confidence intervals)
 justified and discussed, in context, potential problems when reporting on a study and provided examples to illustrate these
 used statistical terms confidently, such as nonsampling error, bias, sampling variability, confounding variables, showing a clear understanding of the meaning of these terms
 were succinct in their responses and presented clear, wellsupported discussion points rather than lengthy generic statements or descriptions
 integrated statistical and contextual knowledge in their responses when discussing issues with a statistical process or claim presented in the statistical report.
Standardspecific comments
Candidates generally demonstrated sound skills with constructing and interpreting confidence intervals. However, some candidates incorrectly made an inference about the respondents rather than the population, e.g. “I’m pretty sure that somewhere between 55% to 89% of respondents”, while others wrote the confidence intervals incorrectly, e.g. 1.7 ± 59; (60.7%, 57.3%); 57.3%  60.7%.
Despite some improvement in the construction of confidence intervals, understanding of sampling variability was not well demonstrated by some candidates. At this level, by quantifying the margin of error using the 1/root(n) rule of thumb, candidates should focus on the size of the sample and its effect on the expected variation of sample estimates.
Some candidates struggled to integrate statistical and contextual information when assessing the quality of reports with respect to conclusions made in those reports. Successful candidates looked for the “obvious” issues in the study design (including the survey design) rather than suggesting obscure issues. For example, in question 2(c), the “obvious” issue was the information about sugar provided to participants before they completed the survey. These issues were then linked to the claims or conclusions made in the statistical report.
Candidates were required to assess the quality of reports with respect to conclusions made in those reports, through questions that scaffolded the evaluation of the statistical report. Candidates needed to read the report and the questions carefully, and highlight or underline key words to focus their responses. Often candidates gave responses that were unrelated to the question and appeared to be rotelearned. Good responses communicate key statistical ideas well, focus on a central idea or point, and support the idea or point with relevant features from the report, e.g. relevant contextual knowledge, specific study design features, and specific results from the study.
For example, in question 3(d), candidates were asked to discuss one potential issue with the study design used for the 2008 survey, in respect to the heading for this report “Higher fines discourage disability parking abuse”. Candidates linked statistical and contextual knowledge by discussing issues such as the presence of confounding variables due to the study being observational, or issues with the sampling of disability parks in terms of representativeness (selection bias). However, this discussion did not address the heading for the report, which was the main conclusion made by the report. To demonstrate statistical insight, candidates need to integrate statistical and contextual information to assess the quality of reports with respect to the conclusions made.
91585: Apply probability concepts in solving problems
Candidates who were awarded Achievement commonly:
 identified relevant probability concepts and contextual information from the problem
 used representations such as twoway tables, probability trees and Venn diagrams
 used methods associated with twoway tables and probability trees
 used methods related to independence to calculate probabilities
 used methods related to conditional probabilities
 demonstrated understanding of true probability versus experimental estimates, e.g. that a sample proportion / relative frequency only gives an estimate for the true probability in the situation
 demonstrated understanding of experimental estimates versus model estimates, e.g. creating a probability model from the results of only one sample, which highlights consequences of nonrandom selection.
Candidates who were assessed as Not Achieved commonly:
 struggled to identify relevant probability concepts and contextual information from the written problem
 used probabilities lower than 0 or greater than 1
 were unable to work with probabilities written in different forms
 did not understand that databased estimates of true probability are subject to sampling variability / sample size and issues of generalisability
 used incorrect methods to calculate probabilities of independent, combined and conditional events
 compared probabilities using differences rather than quotients.
Candidates who were awarded Achievement with Merit commonly:
 selected and carried out a logical sequence of steps, e.g. knowing when to add and when to multiply probabilities when combining events
 demonstrated understanding of the concept of independence, e.g. that for independent events P(B/A) = P(B/A’)
 connected different concepts and representations, e.g. calculating and comparing conditional probabilities to determine “how many times as likely”, using the relationship between probabilities and expected numbers with probability information from a twoway table
 communicated thinking using appropriate statements, e.g. correct use of probability notation.
Candidates who were awarded Achievement with Excellence commonly:
 developed a chain of logical reasoning, e.g. completing a 3event Venn diagram to represent a situation and interpreting the results to solve a problem
 devised a strategy to solve a problem, e.g. discussing how a simulation could be used to consider sampling variability when using experimental estimates of probabilities
 used contextual knowledge to reflect on their answers, e.g. assumptions of independence, assumptions related to sampling without replacement.
Standardspecific comments
When using experimental estimates of probabilities, candidates needed to consider sample size, sampling variability and issues of generalisability. Many candidates incorrectly stated that sample sizes must be equal when comparing groups, e.g. in question 1(a)(iii). Many candidates also lacked familiarity regarding the use of simulations with probability models, e.g. in question 2(b)(ii).
Some candidates used the term “relative risk” when comparing two conditional probabilities using division, e.g. in question 1(a)(ii). The term “relative risk” should be limited to situations where risk is involved. Many candidates also struggled to discuss the concept of statistical independence, e.g. in question 1(b)(i). When events A and B are independent, then knowing that event A has occurred (or not) gives no new information about the chances of event B occurring, i.e. P(B  A) = P(B).
91586: Apply probability distributions in solving problems
Candidates who were awarded Achievement commonly:
 sketched probability distribution models
 described features of probability distribution models
 selected appropriate probability distribution models
 calculated simple probabilities using probability distribution models
 used terms correctly e.g. “more than”, “less than”
 used discrete random variables represented as probability distribution tables
 calculated the expected value (mean) of a discrete random variable
 understood the distribution of true probabilities versus the distribution of model estimates of probabilities, e.g. identifying a factor that should be considered when modelling a situation.
Candidates who were assessed as Not Achieved commonly:
 used incorrect methods, e.g. the wrong choice of model, incorrect probability calculations
 did not account for the standard deviation when sketching a normal distribution model
 misinterpreted instructions such as “fewer than” and “more than”
 misinterpreted features of a probability distribution graph or table.
Candidates who were awarded Achievement with Merit commonly:
 linked features of the probability distribution model to the context, e.g. identifying appropriate assumptions in context, giving the conditions for a probability distribution in context
 combined correctly probabilities and correctly interpreted “OR”
 considered the nature of the variable involved when calculating the probability
 used inverse methods to find unknown parameters of probability distribution models, e.g. inverse Poisson to determine lambda, inverse normal to determine standard deviation
 understood variation from a graphical representation of a probability distribution
 communicated their thinking using appropriate statements, e.g. stating probability distribution model and parameters, correctly using probability notation.
Candidates who were awarded Achievement with Excellence commonly:
 discussed the appropriateness of a model by considering features of the probability distribution, statistical evidence and / or the context
 utilised the relationship between parameters of the probability distribution model and probability (as area under the curve) to compare two different probability distribution models
 interpreted and attempted to generalise features of an experimental probability distribution when considering the features of a discrete random variable based on this experimental probability distribution
 clearly explained their reasoning.
Standardspecific comments
Many candidates struggled to connect the concept of variation, as measured by standard deviation, with the visual features or properties of a probability distribution graph or sketch. For example, in question 1, candidates needed to sketch the probability distribution model (a normal distribution with given parameters) but many sketches did not account for standard deviation. Similarly, in question 3, candidates needed to identify which distribution had less variation and explain why using features of the probability distribution.
Candidates needed to be able to discuss situations and data collected from situations where the mathematical probability distributions (e.g. Poisson, normal, binomial, uniform, triangular) were not appropriate. For example, when answering question 3(b)(ii), many candidates struggled to explain why it would not be appropriate to model the ratings for the ‘before’ survey using the conditions for a Poisson distribution.