Marking Criteria

1
Marking Criteria

• Correct application of statistical tests
• Presentation
• Interpretation
• Communication of information in response to the
  data and research questions
• 40 of total mark for course

2
Format of assignment

• Report the findings in the form of an academic
  paper
• Abstract
• Introduction
• Methods
• Results
• Discussion and conclusion
• 2000 word limit

3
Abstract/introduction

• Abstract
• Communicates summary of report
• Structured or unstructured
• Incorporates all key features
• Introduction
• Introduce research question, the aim of the
  research, specific objectives
• No requirement for extra reading
• Enough information for someone reading the paper
  (who has never seen the data) to appreciate why
  the analyses has been conducted

4
Methods

• Detail how and what data has been collected and
  from where (info given in assignment)
• Describe statistical methods that have been
  applied
• Detail any assumptions and how these have been
  checked
• Justification for statistical methods applied
• Data manipulation i.e. recoding of variables
  again with some justification
• Enough information for someone to redo your
  analyses

5
Results 1

• Consider the use of tables, graphs, text
• What summary/descriptive information is required?
• Informative presentation of data
• Description of sample
• Comparison of groups
• Association of variables
• In the assessment of assumptions, there is no
  requirement to produce graphs of variables, a
  comment in the text will be sufficient

6
Results 2

• Present 95 CIs where appropriate
• Appropriate use of output provided in SPSS
• Consider what information from SPSS is required
  to communicate to the reader the inference you
  are making about the population
• DO NOT CUT AND PASTE LARGE AMOUNTS FROM SPSS
• Extract only what is required and present in
  appropriate tables, text etc
• Graphs can be taken directly from SPSS

7
Discussion and conclusion

• Discuss analysis and results found
• Unusual findings
• Consistency of results
• Strengths and limitations of study to address
  research question
• Any additional information
• Other statistical methods that could have been
  applied
• Based on the design and findings, are the results
  conclusive?
• What do the results from this study inform us
  about the factors associated with greater risks
  of death in the ICU in the population of people
  treated in ICU?

8
What to do next when writing a real report, paper
or thesis

• Look at published papers
• Put your results in context
• Compare and contrast with similar national or
  international studies
• Consider whether differences or limitations in
  your own and other study designs could explain
  contrasting findings
• Your conclusions will be dominated by your own
  findings but must take account of other work

9
Null hypotheses

• For tests comparing means paired t, independent
  t, ANOVA
• The null hypothesis is about the MEANS in the
  population
• Paired t The mean of the differences is equal to
  zero in the population
• The mean change in weight 0 in the popn.
• Independent t The mean weights in the two groups
  are equal in the population
• ANOVA The mean weights in the 3 (or more) groups
  are equal in the population

10
Null hypotheses

• For tests comparing distributions Wilcoxon,
  Mann-Whitney, Kruskal-Wallis
• The null hypothesis is about the DISTRIBUTIONS in
  the population
• Wilcoxon The distribution of the differences is
  centred on zero in the population
• The was no change in the distribution of weights
  between the two time points (before vs. after) in
  the population
• Mann-Whitney The distributions of the weights in
  the two groups are equal in the population
• Kruskal-Wallis The distributions of in the 3
  (or more) groups are equal in the population

11
Quoting

• P-values
• give the actual value
• particularly important if significant
• NEVER WRITE P 0
• INSTEAD WRITE P lt 0.001
• Non-parametric tests
• Use Medians and IQR to give context
• (and to help you understand in which group are
  the measurements higher)
• DONT QUOTE MEAN RANKS

12
Significance and confidence intervals

• p ? 0.05 was taken to be statistically
  significant.
• The result was significant at the 5 level.
• The result was significant at the 1 level.
• The difference in means was statistically
  significant (P0.03).
• Smokers were significantly more likely to have an
  MI than non-smokers. The p-value was 0.021.
• A 95 confidence interval for the difference in
  the population means is presented in the table.
• NOT
• The results were significant at the 95
  confidence level.
• The results were significant at the 95
  confidence interval.

13
Percentages

• Out of what group is it a percentage or
  proportion?
• Suppose there 189 women in total. 96 of these
  women are white and 23 of these white women have
  low birth weight (LBW) babies
• The proportion of white women having low birth
  weight babies is 23 / 96 0.24. Percentage24
• The proportion of low birth weight babies whose
  mothers are white is different.
• The proportion of the sample who were white women
  with LBW babies 23 / 189 0.12 Percentage12
• Comparisons cannot be done with TOTAL
• Use ROW or COLUMN for comparisons.
• Dont mix and match when reporting on a single
  table.
• 41 of smokers and 25 of non-smokers had a low
  birth weight baby. CLEAR MESSAGE.