Discuss the importance of assessing the appropriateness of the statistical tests

Discuss the importance of assessing the appropriateness of the statistical tests

Discuss the importance of assessing the appropriateness of the statistical tests

January?2?::?vol?27?no?18?::?2013? © NURSING STANDARD / RCN PUBLISHING

Learning zone CO N T IN UIN G  P R O F E SSIO N A L D E V E LO P M E N T

Understanding quantitative research: part 2 NS674 Hoare Z, Hoe J (2012) Understanding quantitative research: part 2. Nursing Standard. 27, 18, 48-55. Date of submission: July 1 2011; date of acceptance: March 2 2012.

Abstract

This?article,?which?is?the?second?in?a?two-part?series,?provides?an? introduction?to?understanding?quantitative?research,?basic?statistics?and? terminology?used?in?research?articles.?Understanding?statistical?analysis? will?ensure?that?nurses?can?assess?the?credibility?and?significance?of?the? evidence?reported.?This?article?focuses?on?explaining?common?statistical? terms?and?the?presentation?of?statistical?data?in?quantitative?research.

Authors Zoë Hoare Clinical?trials?statistician,?Bangor?University,?Bangor. Juanita Hoe Senior?clinical?research?associate,?Research?Department?of?Mental?Health? Sciences,?University?College?London,?London. Correspondence to: [email protected]

Keywords Data interpretation, parametric statistical tests, quantitative research, statistics.

Review All?articles?are?subject?to?external?double-blind?peer?review?and?checked? for?plagiarism?using?automated?software.

Online Guidelines?on?writing?for?publication?are?available?at www.nursing-standard.co.uk.?For?related?articles?visit?the?archive?and? search?using?the?keywords?above.

4 Page 57 Statistics multiple choice questionnaire

4 Page 58 Read Sarah Holling’s practice profile on head injury

4 Page 59 Guidelines on how to write a practice profile

Aims and intended learning outcomes This article aims to provide a useful introduction to common statistical terms and the presentation of statistics in research articles. After reading this article and completing the time out activities you should be able to:

Discuss the importance of assessing the appropriateness of the statistical tests performed and accurate interpretation of findings.

Recognise the common statistical tests used in quantitative research.

Identify errors in the reporting of statistical analysis, such as selective reporting and overestimating the significance of findings.

Understand the importance of statistics in evidence-based knowledge relevant to your area.

Introduction Statistics are the methods and techniques used to collect, analyse, interpret and present data (Maltby et al 2007). Nurses routinely use statistics within their practice, for example when they give health information to patients about their diagnosis or prognosis and in discussing the adverse effects of medication or treatment. However, many nurses may find understanding the presentation of statistical data within a research article challenging.

Fear of statistics is common and is usually linked with anxiety about understanding and interpreting statistical data and outcomes (Williams 2010). In health research, statistics may be used to determine the prevalence and incidence of illness or establish if a new treatment is effective.

© NURSING STANDARD / RCN PUBLISHING january?2?::?vol?27?no?18?::?2013? 49

Results from the statistical analysis are key in establishing the evidence. They must also be examined carefully to ensure that data collected are presented and interpreted accurately. It is important to observe whether results are misleading, for example if there is evidence of selective reporting or the significance of findings is overestimated. It should be noted that statistically significant findings are not always clinically significant (McCluskey and Lalkhen 2007).

Reliable research evidence will, however, support the implementation of evidence-based interventions in practice. By contrast, weak evidence may indicate a need for further research. A knowledge of basic statistics is therefore essential and will help nurses to understand and assess the credibility and significance of the evidence presented.

Interpreting data Descriptive data, also known as , is information provided about the sample population. This data usually includes the sample size and demographic characteristics, which are either described by frequencies (the number of observations) for categorical variables, such as gender and ethnicity; or by the mean (average) number and standard deviation (measure of variance) for continuous variables, such as age or years of education (Table 1).

Ranges that show the lowest and highest measures within that sample should also be provided; for example, the range for age will show the youngest and oldest ages within the sample. Where there are two sample groups, such as the treatment and control group, it is important to look for similarities between the two groups to ensure they are comparable. If the mean scores and the range of the measures obtained vary significantly between the two groups, the samples may not be considered comparable and this would introduce bias (prejudice) into the results of the trial.

It is important to understand the characteristics of the sample within the study, as this is the population to which the results apply and will determine whether they are generalisable to the wider population for that patient group. However, caution is needed when generalising results; for example, the results of a study undertaken in the United States (US) cannot be generalised to the UK population. Although there may be similarities between the two populations, cultural differences

exist. Therefore, the study would need to be replicated in the UK to see if similar results are recorded in this population. Complete time out activity 1

Presenting data

When examining data presented in tables it is worth considering how the information has been collected. This is because gender and age are not subjective measures, but other recorded outcome measures may be, such as participants’ mood or behaviour, which may change over time. For example, the Beck Depression Inventory (Beck et al 1961) measures the severity of depression in a person.

The measurement of the tool is subjective, meaning that scores will vary for an individual over time depending on how he or she is feeling. However, if the tool has been validated correctly then the stability of the measure should have been assessed through thorough examination of content, comparisons and factor analysis. The Beck Depression Inventory is considered a valid and reliable measure of depression and is widely used across different population groups.

Using the Body Mass Index (BMI) (Keys et al 1972) within a population, one would expect to find a few very low BMI scores and a few very high BMI scores, but most would be centred around the mean score. The mean score can be affected by the extreme values (outliers) making it higher or lower than expected. For example, mean income can be affected by a few highly paid workers even though the majority are on a much lower wage. For this reason, social scientists tend to use the median (middle value) when describing UK household income.

Understanding the spread of data is important.

Figure 1 shows a standard normal distribution (spread or shape) in red where the mean score is zero, and the middle of the distribution and standard deviation is one, meaning that about 68% of the sample have a value within ±1 of the mean. The two other distributions shown are skewed (not equally distributed around the mean), where the mean of the distribution is not necessarily the middle – the means are ±4 in this case.

The blue line is a positive right skew, where more cases are to the left of the distribution. It is right skewed as the tail extends out further to the right than expected. The green line is a negative left skew with more cases to the right of the distribution and a longer tail out to the left of the distribution.

1 Locate a quantitative research article with figures and tables used to represent data from a randomised controlled trial. Read the article and examine the descriptive data presented for the intervention and control samples. Look for similarities within the samples; for example, are the proportion of males and females in the two groups the same? Are the two groups of similar age and range of ages? Do the two samples have similar mean scores for the measures used, for example mood, dependency or quality of life? Are there any major differences in mean scores between the two groups, indicating that the sample populations are not matched?

50? january?2?::?vol?27?no?18?::?2013? © NURSING STANDARD / RCN PUBLISHING

Learning zone statistics

Parametric statistical tests typically assume that the distribution of values takes the same shape as the red line in Figure 1, and care should be taken when interpreting the results of the research if another shape is apparent.

If the distribution of data does not follow that of the standard normal curve, then applying the commonly used parametric tests could cause the incorrect inferences to be drawn.

The common statistical tests used to analyse data are typically parametric tests. They usually assume that data are normally distributed and have more statistical power than non-parametric tests, which are used when there is no assumption that data are normally distributed (Greenhalgh 1997). Statistical signifi cance is more diffi cult to show with non-parametric tests (Greenhalgh 1997). Complete time out activity 2

Hypothesis testing and statistical significance

A statistical hypothesis is an assumption about a population parameter (value), which the study will test. This assumption may or may not be true. The null hypothesis assumes that changes to the sample result from chance and that there is no difference between the two test scores or there is no difference from zero. The alternate hypothesis assumes that changes are infl uenced by some non-random cause. The alternate hypothesis states there

TABLE 1 Common statistical terms

Statistical term Description

Mean To?calculate?the?mean?(average)?score,?take?all?the?values,?add?them?up?and?divide?the?total?by?the?number?of? values.?The?mean?score?is?often?thought?of?as?the?middle?of?a?distribution,?however?this?is?only?true?when?the? distribution?takes?certain?shapes,?for?example?the?standard?normal?distribution?curve?(Figure?1).

Median This?is?the?middle?value?of?the?distribution.?To?calculate?the?median,?line?all?the?values?up?smallest?to?largest.? For?an?odd?number?of?values?the?median?becomes?the?middle?value.?For?an?even?number?of?items?the?median? becomes?the?mean?of?the?two?central?values.

Probability The?probability?is?the?number?of?times?an?event?occurred?divided?by?the?total?number?of?times?the?event?was? attempted.?A?probability?value?will?always?be?between?0?and?1.?A?value?of?0?means?that?the?event?never?occurs? and?1?means?that?the?event?always?occurs.?This?is?reported?as?the?P?value.