In this section we provide information about useful tools for designing and analyzing experiments – instruments and equipment, which we at PAASP frequently use. In addition, we also introduce helpful resources for educational and training purposes.
The concept of block randomization and a free-ware online tool
In a previous post we discussed the concept of ‘simple randomization’; now we would like to introduce ‘block randomization’ as a really helpful tool to prevent bias in research experiments:
Simple randomization, as the easiest way to assign samples to different groups, has its disadvantages when sample sizes are small. For example, when animals need to be selected from different cages, simple randomization could, with great probability, result in an imbalance between treatment groups in case there was an environmental influence that differs between cages, which then could introduce bias. To avoid this, animals should be “randomly” selected into blocks, represented by the number of cages and animals from (ideally) just one treatment group.
As an illustration: 20 animals will be treated during an experiment with 4 different conditions and are therefore supposed to be separated in 4 groups of 5 animals each. These animals are housed in cages with 5 creatures. In this case, it is necessary to ensure that each group contains at least one animal from each cage so that an evenly balanced distribution of animals is guaranteed. However, within one block, the distribution of the different treatment types should be random. This block randomization is a common technique used in clinical trials to assure that equal numbers of patients are treated with placebo and active compound at each research facility. Therefore, this block randomization step will make sure that the ratio between placebo and treatment groups is comparable at all research facility. Like this it can be excluded that potential biases are introduced due to an unintentional favoring of one group at a specific facility.
Before presenting available tools for block randomization, two more examples from in vitro experiments:
First, when working with oocytes, their conditions and characteristics will most likely change depending on their time of collection over a one day period and they will appear and behave slightly different in the afternoon compared to the morning. When performing 6 different treatments in triplicates, it would therefore make sense to have three blocks, each comprising one of the triplicates, and randomly assign each treatment within each block. Like this it is possible to ensure equal distribution of treatment groups over the whole day and sufficient variations within the blocks.
Second, it is even possible to use block randomization when analyzing tissue samples for protein expression by Western blotting. Given the previous example of 20 animals in 4 groups, to ensure that there is no experimental artefact due to the order in which these tissue samples are loaded on the SDS-PAGE gel, the groups can be transformed into five blocks with one sample from each treatment group in each block. This will guarantee that the 20 samples from all different animals are evenly distributed over the gel and no technical bias can occur.
There are several online tools for block randomization, however, all of them have their advantages and limitations. Here, we would like to introduce ‘sealed-envelope’ in more detail (link) using the last scenario as an example:
A screenshot of the web page ‘sealed-envelope’ is presented below. As a first step, the different treatment groups (e.g. Group A – D) and the block sizes (e.g. 4) should be designed. For the latter, however, you can only choose multiples of your treatment groups. Given the 20 animals used in the experiment, the List length is 20.
Next, creating a list will result in 5 different blocks with 4 randomly allocated samples each (see screenshot below). This can be used to define the order on how to load the samples on the gel, therefore, providing basic block randomization. Going back to the first example with only 4 blocks, the last group will have to be distributed to the first 4, which can be easily done by adding ‘5, 4, 1, Group C’ to the first block identifier, ‘5, 4, 2, Group B’ to block identifier 2 and so on. This will provide four groups with unbiased selection of the animals.
For the oocytes example, unfortunately it is not possible to have 6 treatment groups and 3 block sizes as needed for this experiment. In this particular case, it will be necessary to use a simple randomization tool to randomize the six treatment groups and repeat it three times. This will give the sequence of treatments for the whole day and guarantee best and unbiased distribution.
However, this seems to be an unsatisfying process and since we are not aware of any tool to easily allocate these treatment groups in blocks when planning individual experiments, PAASP is developing a functional solution for this issue in collaboration with a software IT company. We will keep you informed about our progress on this topic.
The next randomization strategy we will discuss in one of the upcoming newsletters is called ‘stratified randomization’.
Block randomization: R code
There are a variety of free online tools that can be easily accessed to address various data analysis needs, such as the randomization tool described above. One would only need to be aware of such resources, maintain these references and links in a convenient format and place and, last but not least, make sure that these resources are available in the future when the need comes.
Looking for something free, good-quality, easily accessible and likely to be always there for you? There is such a tool: ‘R’, a data analysis and graphics package that is becoming increasingly popular as it is seen by many people as a tool to establish reproducible research practices (as explained in this Coursera course – link as well as in many publications found online).
The graphics part of R may seem difficult to use, but writing the data analysis scripts can be quickly learned as there are numerous free examples shared by the vast R community (e.g. statmethods or r-bloggers). You may want to read this introduction (computerworld), that contains a number of useful references, or look at this Tutorial link.
After you have installed R (using ‘R Studio’ to make communication with R easier), you will be able to ask Google or the R community (see the links above) a question of interest (e.g. how does one do ANOVA in R?) to get clear instructions.
R has specialized packages for a variety of specific tasks. For Block Randomization, a special R package is available (it takes less than a minute to install) and a complete manual can be found here: r-project.
If you find it difficult and need more time to learn, we are certain your colleagues at the biostatistics departments are skilled in R programming and will be happy to provide you with a ready-to-use script.