Factors affecting choice of statistical tests, including level of measurement and design Flashcards
What are descriptive statistics
Descriptive statistics can give summaries of data that we have collected from our research (tables, bar charts etc) and indications of what statistical analysis may reveal about our results.
What are the three levels of measurement and what is a level of measurement?
Levels of measurement are used to try and categorise our data into one of three types, allowing us to correctly select the most appropriate statistical test to analsyse our results.
The three types are nominal, ordinal and interval.
Nominal data
Nominal data can be referred to as categorical data, in nominal data there are distinct categories. ie students ages would either be 16 or 17, and if the data is nominal then each participant will only appear in one category. This is called discrete data. In the example above you are either 16 or 17 you cannot be both.
Ordinal data
Data is ordinal if it is ordered in some way and the intervals between the data are not equal. Typically used to simply rank data where the values assigned have no meaning beyond the purpose of stating where one score appeared in relation to others.
An example is people were asked to rate a group of restaurants from 1-10 with one being their least favorite and ten being their most favorite. They could then rank the restaurants based on their average score. But they would not be able to say if the difference between 1st and 2nd place is equal to that from 2nd to third.
Interval data
Interval data like ordinal data refers to data that is ordered in some way however the intervals between each value are equal in measurement. This data is often more objective and scientific then ordinal data due to this.
Examples of types of interval data are temperature and time. the difference between 1 Celsius and 2 Celsius is the same as the difference between 32 Celsius and 33 Celsius. Temperature also has a set numerical order.
Evaluation of nominal data
+) Nominal data is easily gathered from closed questions on a questionnaire or interview. This can often be generated quickly and tested in timely manner increasing reliability. The mode is measure of central tendency that can be applied to nominal data.
-) There is no scale of reference for nominal data often making the data seem overly simplistic. There is also no measure of dispersion/spread that can be used on nominal data.
Evaluation of ordinal data
+) Ordinal data provides more detail than nominal data as the scores are ordered linearly.
-) The intervals between scores are not of equal value, this means that the mean can’t be used as a measure of central tendency. This often leads to the median being used instead.
Evaluation of Interval data
+) Interval data is considered more informative than the nominal and ordinal levels of measurement. The gaps between scores are of equal value increasing reliability.
-) In some instances the intervals can be arbitrary. For example 100 degrees Celsius is not twice as warm as 50 degrees Celsius. However other more scientific forms of measurement such as Kelvin can be used instead.
What are the 8 statistical tests?
Non-parametric:
1) Chi-sqaured
2) Spearman’s Rho
3) Man Whitney
4) Wilcoxon
5) Sign test
Parametric:
6) Pearson’s r
7) Related t-test
8) Unrelated t-test
Parametric tests vs non parametric tests
Parametric tests are more robust and powerful than non parametric tests, They rely on the actual data collected rather then just examining the rank order of the data. Parametric tests are also more likely to detect if the data is significant or not.
However there are 3 factors that allow for a parametric tests to be conducted.
1) The data must be interval it cannot be ordinal or nominal.
2) The data should be taken from a population with a normal distribution curve rather then a skewed one.
3) The data should have similar variance or spread of scores. This can be examined by looking at the dispersion of the data and standard deviations for both conditions to see if they are similar.
These restrictions are not in place for non- parametric tests.
How do i decide what test to use?
1) Is it a test of difference, correlation or an association?
2) What research design is being used? Independent measures, repeated measures or matched pairs design?
3) Which level of measurement is being used in the research? Nominal, ordinal or interval?
Look at the table on pg 42 and learn that. Remember: Carrots Should Come Mashed With Swede Under Roast Potatoes
