I've been reading up on how to best represent qPCR data, specially on answers given on this site by Dr. Jochen Wilhelm. I understand box plots are the way to go, but I still don't understand them completely.
For a box plot on qPCR data, what should I use as the range and why? Should I use the Q1-Q3 as range? I've also been looking at different ways to represent the error, and I've realized that when I use Q1-Q3 as range and standard error as the error, the error bars are on the inside of the box, and I don't understand what this means.
Plus, when representing standard error, what coeficient should I use and what does the coeficient mean? I've noticed the error bars change depending on the coeficient used for the standard error, but I also don't understand what this means.
If anyone could help me understand these concepts and the best way to represent this kind of data, I'd greatly appreciate it.
There is a standard way how to create boxplots, see e.g. here:
https://towardsdatascience.com/understanding-boxplots-5e2df7bcbd51
(there is a Wikipedia article on boxplots, but the English article is a bit weird; the German article is quite ok, but well... it's in German; the Spanish varticle is short but seems to be on the point: https://es.wikipedia.org/wiki/Diagrama_de_caja)
The box is usually showing the IQR (=range between Q1 and Q3). If not, it has to be clearly stated to avoid confusion (I'd advice against using anything else but the IQR as the box). The whiskers are usually extended up to the most extreme value within a distance of 1.5xIQR from the box. Some authors extend the whiskers to the most extreme values (it may coincide that the most extreme values are withing that 1.5xIQR distance). Again, this should be clearly stated, as this is not quite the standard way to draw the whiskers. And if there are values outside the whiskers, they are typically drawn as individual points. They might be omitted, but that, again, should be clearly stated (one would miss extreme outliers that can cause problems in the further analysis/interpretation and that should be investigated).
Boxplots are used to show the distribution of the values based on the empirical quantiles. They should not show anything different. They are not meant to show parametric estimates (means, standard deviations, standard errors, or confidence intervals).
Boxplots (in their standard form, as described e.g. in the link) are a good way to show Ct and dCt values. If ddCt values are calculated from paired samples, they can also be shown as boxplots. If ddCt values are results of comparisons of unpaired groups, these are single values, which can be shown as single values (together with their standard error or confidence interval) - there is no point in using boxplots in this case.
Note: Boxplots actually show 5 statistics from the data (lower whiker end, Q1, Q2, Q3, and upper whisker end). It makes no sense to "summarize" just a hand full of values by as many as 5 statistics. I have seen boxplots being made from as few as 3 or 4 values per group. This is complete rubbish! Boxplots are useful and valuable when you have at least 20 values per group (but even then I advise to consider just showing all the individual values rather than some summary). In biological research is is rarely the case to have that large sample sizes that boxplots would make sense (better just show the values!).
There is a standard way how to create boxplots, see e.g. here:
https://towardsdatascience.com/understanding-boxplots-5e2df7bcbd51
(there is a Wikipedia article on boxplots, but the English article is a bit weird; the German article is quite ok, but well... it's in German; the Spanish varticle is short but seems to be on the point: https://es.wikipedia.org/wiki/Diagrama_de_caja)
The box is usually showing the IQR (=range between Q1 and Q3). If not, it has to be clearly stated to avoid confusion (I'd advice against using anything else but the IQR as the box). The whiskers are usually extended up to the most extreme value within a distance of 1.5xIQR from the box. Some authors extend the whiskers to the most extreme values (it may coincide that the most extreme values are withing that 1.5xIQR distance). Again, this should be clearly stated, as this is not quite the standard way to draw the whiskers. And if there are values outside the whiskers, they are typically drawn as individual points. They might be omitted, but that, again, should be clearly stated (one would miss extreme outliers that can cause problems in the further analysis/interpretation and that should be investigated).
Boxplots are used to show the distribution of the values based on the empirical quantiles. They should not show anything different. They are not meant to show parametric estimates (means, standard deviations, standard errors, or confidence intervals).
Boxplots (in their standard form, as described e.g. in the link) are a good way to show Ct and dCt values. If ddCt values are calculated from paired samples, they can also be shown as boxplots. If ddCt values are results of comparisons of unpaired groups, these are single values, which can be shown as single values (together with their standard error or confidence interval) - there is no point in using boxplots in this case.
Note: Boxplots actually show 5 statistics from the data (lower whiker end, Q1, Q2, Q3, and upper whisker end). It makes no sense to "summarize" just a hand full of values by as many as 5 statistics. I have seen boxplots being made from as few as 3 or 4 values per group. This is complete rubbish! Boxplots are useful and valuable when you have at least 20 values per group (but even then I advise to consider just showing all the individual values rather than some summary). In biological research is is rarely the case to have that large sample sizes that boxplots would make sense (better just show the values!).
I use graphpad (PRISM) to create dot or box plots (then use Q1 and Q3) and is a convenient way to analyse your data. Advantage of a dot plot with standard deviation is the readers can see the distribution of your data points and assess if the data is corresponding to the observation and with a box plot you will lose that in some way.
- Badges
- Science method