Home
Subjects
Explanations
Create
Study sets, textbooks, questions
Log in
Sign up
Upgrade to remove ads
Only $35.99/year
Statistics Lane and Sandor 2005 Reading
STUDY
Flashcards
Learn
Write
Spell
Test
PLAY
Match
Gravity
Terms in this set (13)
Why should we include graphs in our research?
Graphs are better at summarizing data, showing trends, and showing points and patterns, whereas tables are better for point/value reading
Easy to perceive the overall pattern of results
What are the four types of errors that come with displaying graphs?
Often poorly constructed in science
Four types of errors: construction errors, degraded image errors, errors in the explanation, and discrimination errors
What should be portrayed in graphs?
Average published graphic had a low data density
How to portray three types of information in graphs: shapes of distribution, trends, and inferential statistics
Distributional information about graphs: What is a common mistake?
lack of information regarding the shape or distribution of the data and that this lack of info hinders scientific evaluation
Majority of graphs were bar charts, and only 10% of the graphs showed distribution information beyond central tendency
Example: if you want to look at three conditions and two groups and you use a bar graph, it would contain no information about the variability or shape of the distributions
One possible objection to including distributional information is that the info may distract readers from what is typically the main objective of the graph: portraying the pattern of means
Design solution is preferable to a content reduction
Some researchers may resist showing distributional data that reveal the variability in a plot of means
An Effective Display of Data MUST:
Reveal the uncertainty in the data
Characterize the uncertainty as it relates to inferences to be made from the data
Help prevent the drawing of incorrect conclusions due to lack of appreciation of the precision of the information conveyed
Standard error bars are not often sufficient enough for communicating inferential information
Although easy to interpret, we believe that graphs specifying the means that are significantly different from each other have several drawbacks:
Encourages the all-or-none rejection of a null hypothesis
Emphasizes hypothesis testing and neglects confidence intervals
May distract visually from the pattern of means if several means are being compared
Important to consider how the target audience understands the graphs
Cumming and Finch (2005) proposed ways to design graphs so that they would show inferential statistics in a more meaningful way
They gave examples of how graphs could display CIs relevant to inference in question
Example: in a two-condition experiment, one should include a graph showing the difference between means and CI on this difference in addition to the group means and their respective CIs
Figure 9
DO NOT DISPLAY YOUR P VALUE IN A BAR GRAPH
Why shouldnt you display your p value on a graph?
Remember, some inferential statistics (I.e., p values) is better to present in text rather than graphs
A graph only showing means does not provide sufficient information for the reader to know how seriously to take sample differences or to judge the likely size of the difference in the population
Solution: refer reader to text for inferential statistic
What is wrong with the The Archaic Separation of Graphics and Text?
Tufte argued that the separation of graphics and text is unnecessary as it suppose to bring together virtual, verbal, and quantitative information together
These new journals require readers to skip back and forth between figures, tables, etc
Cognitive demands for reading the graphs is lower when various modes of info are displayed together in an integrated format
Using a graph, like box plots, to portray group differences has two potential problems when used in an ANCOVA design:
(1) Because inferential tests are done on differences among adjusted means (controlling for the covariate), differences in means portrayed in the graph would not reflect the differences tested in ANCOVA
(2) The variability of the distribution shown in the graph would include the variability potentially controlled by the covariate and would therefore be greater than it should be
Solution: remove the effect of the covariate from the data before creating the box plots
Whats wrong with bar charts showing group means?
Important to dissect graphs and understand what they mean
Bar chart showing group means lack distributional data
So, how do we integrate text and graph?
Create the statistical graph with one of the many widely available statistics packages
What are the Principles for Constructing Good Graphs?
Well-constructed graphs help readers focus on important aspects of data, provide visual clarity, and make interpretation easy
One should avoid using shaded backgrounds that reduce contrast, eliminate unnecessary elements, and use sufficiently dark lines and points
Careful to avoid apparent inconsistencies between the graph and the results of the statistical analysis
Example: in within-subjects design, the graph should control for variation due to participants, just as statistical analysis would
Designed to decrease mental load
Important for consistency of assignments of symbols in figures
Place graphical elements that are likely to be compared near each other
Too much info contained the graph makes it difficult to interpret
Graphs can mislead rather inform
Use of double y axis can be misleading
Other sets by this creator
Statistics Lecture 4: The t-Test Lecture
15 terms
Statistics: Greenland (2016) Article
5 terms
Statistics- Cumming and Finch (2005)
14 terms
Statistics Midterm: Distribution of Sample Means a…
45 terms