lecture #4 / chapter #5 - what visualizations do for research

what visualizations do for research

• graph makes patterns in datat visble

  • does not explain why patterns occurred

• choose a display that matches the question

• describe a distribution

• useful figure → see what observations and uncertainty without hiding important variation

data visualization is part of statistical reasoning

1. graph should help a reader compare relevant values and detect patterns

2. makes the measurement, groups and scale easy to identify

3. use a graph to explore and communicate evidence

start with questions and variable types

• identify what each varibale measures

  • categorical, nominal, ordinal, etc

• decide whether you need to show a distribution

• chart type follows the design and question - not simply the software menu

principles for clear data (3)

• show measured values faithfully

• remove distraction

• keep enough context for a reader to understand what each mark represents

histograms for quantitative distributions

• histogram groups quantitative scores into adjacent intervals

• touchign bars → intervals cover a continuous scale

• for diagnosis type, use a bar chart w/ seperated bars instead

histograms plot:

• the score (x-axis)

• the frequency (y-axis)

histograms help us to identify:

• shape of distribution

  • skew

  • kurtosis

  • spread variation in scores

• unusual scores

population pyramids compare distributions

• mirrors two distributions around a shared centre line

• the left-right direction is a display convention

• compare location, spread and shape while checking whether both sides use the same scale

read a population pyramid carefully

• emphasizes how many observations fall at each score range in each group

• groups differ in sample size, use proportions or percentages when the goal is to compare shape

• visual differences are dscriptive

boxplots - the boxes show:

• median

• upper and lower quartile

• limits within which the middle 50% of scores lie

boxplots - the whiskers show:

• range of scores

• limits within which the top and bottom 25% of scores lie

what boxplot summarizes

• line inside the box is the median

  • runs from first to third quartile

• box covers the IQR, the middle 50% of scores

• whisker rules vary by software

error bar charts

• bar usually shows the mean score

• error bar sticks out from the bar like a whisker

error bar displays the precision of the mean in one of three ways:

1. confidence interval

2. standard deviation

3. standard error of the mean

error bars: know what they encode

• standard deviation describes variation among individual scores

• standard error describes the sampling variability of an estimate

• a ci gives range of estimates compatible with a stated procedure and confidence level

different bar-chart formats answer different questions

• simple bar chart summarises one categorical predictor

• repeated-measures displays should make the within-person structure clear

• for quantitative outcomes, plots showing raw observatoins or distributions oftenr eveal more than bars alone

read mean differences with uncertainty

• bar height represents a group mean

• check how large the difference is in outcome units and how uncertain each estimate is

• error-bar overlap is not a reliable stand-alone signficiance test

read the matrix one pair at a time

• eahc off diagonal panel shows two variables; identify the x and y axis before describing its trend

• check whether a few cases dominate the pattern and whether relationships are linear

• pairwise plots help exploration, but causal conclusions require a design and model that support them