--- title: Distribution shape description: "It's useful to think of distributions of continuous variables in terms of their \"central location\", typically the mean or median; their spread, typically their standard deviation (SD) or inter-quartile range (IQR) and the Gaussian (\"Normal\") distribution is completely defined by…" url: "https://www.psyctc.org/psyctc/glossary2/distribution-shape/" updated: "2021-11-13" category: All OM book glossary entries --- # Distribution shape It's useful to think of distributions of continuous variables in terms of their "central location", typically the mean or median; their spread, typically their standard deviation (SD) or inter-quartile range (IQR) and the Gaussian ("Normal") distribution is completely defined by just those two parameters (mean and SD). However, real world distributions are more complex and it's useful to think of their shape. After central location and spread the next shape issues are skew (roughly assymmetry) and "kurtosis": how "heavy" the tails are. In addition distributions may have multiple lumps: if you ask people to give you numbers between 0 and 100 "at random" they won't be random and often they have lumps at round numbers and sometimes the 5s: 25, 35, ... 95. ##### Implications Thinking about shapes of distributions particularly of therapy change measure scores but also of age does matter and add important information to simply knowing the mean and SD. If a distribution of first session scores for one clinic extends lower than for another because the second service is not accepting into therapy clients scoring below some cut off score but if the first also has a lot of social deprivation it may have markedly more clients starting therapy with high scores and yet the means may be very similar for the two services (but the SD will be larger for the first) and the two services may actually be dealing with really rather diferent challenges. ## Details and examples It's really useful to get familiar with looking at distributions occasionally as density/probability plots (next image) but generally in [histograms or bar graphs](https://www.psyctc.org/psyctc/glossary2/histograms-and-barplots/), [boxplots](https://www.psyctc.org/psyctc/glossary2/boxplot-or-box-plot/) and [violin plots](https://www.psyctc.org/psyctc/glossary2/violin-plot-or-violinplot/) (see entries for more on each of those). Here's a probability/density plot of two theoretical (near) Gaussian distributions differing only in location and spread. (Only "near Gaussian" as the scores can only take values between finite limits, 0 and 20 hear by the look of it; the true Gaussian goes from minus infinity to plus infinity.) ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/Gauss2-1024x885.png) ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/GaussHistOnX-1024x885.png) Histogram of a discrete distribution approximating closely to a Gaussian ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/posSkew-1-1024x885.png) Histogram of data with positive skew (long tail to the right, higher values, hence positive) ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/negSkew-1-1024x885.png) Histogram of data with negative skew (long tail to the left, lower values, hence negative) ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/06/figure_5.8-2048x1280-1-1024x640.png) Histogram with two peaks/modes (Figure 5.8 from the book) ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/GaussBox1-1024x885.png) Boxplot of standard Gaussian distribution, boxplot almost useless for to show shape as shown in ... ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/BoxAndHist001-1024x885.png) Same data as in last plot, boxplot on left, histogram (with Gaussian curve) on right, but ... ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/COREOMBySocStatnBox-1-1024x885.png) Notched boxplot of scores against social status: you can see positive skew and can compare central location ![](https://www.psyctc.org/psyctc/wp-content/uploads/2021/11/geomViolin1-1024x885.png) Violin plot showing some positive skew #### Try also [Distributions](https://www.psyctc.org/psyctc/glossary2/distribution/) [Gaussian ("Normal") distribution](https://www.psyctc.org/psyctc/glossary2/gaussian-normal-distribution/) [Mean](https://www.psyctc.org/psyctc/glossary2/mean-arithmetic-mean-average/) [Standard Deviation (SD)](https://www.psyctc.org/psyctc/glossary2/standard-deviation-sd/) [Variance](https://www.psyctc.org/psyctc/glossary2/variance/) [Median](https://www.psyctc.org/psyctc/glossary2/median/) [Inter-Quartile Range (IQR)](https://www.psyctc.org/psyctc/glossary2/inter-quartile-range-iqr/) [Range](https://www.psyctc.org/psyctc/glossary2/range/) [Minimum](https://www.psyctc.org/psyctc/glossary2/minimum/) [Maximum](https://www.psyctc.org/psyctc/glossary2/maximum/) [Histogram](https://www.psyctc.org/psyctc/glossary2/histograms-and-barplots/) [Boxplot](https://www.psyctc.org/psyctc/glossary2/boxplot-or-box-plot/) (and [notched boxplot](https://www.psyctc.org/psyctc/glossary2/notched-box-plot/)) [Violin plot](https://www.psyctc.org/psyctc/glossary2/violin-plot-or-violinplot/) #### Chapters Touched on in chapter 5 but the principle of "looking at your data" runs through the book. #### Dates Created 13/11/21