Question

As the project team and technical staff move into the design phase, their success will be constrained by how well the analysts have documented the instructions or guidelines. In addition to the program details, the analyst will often provide flow charts and specific details describing the nature of the data.

What type of errors could the analyst make?

Answer 1

Please Upvote, If you are impressed with the answer

If you have any doubt, you can comment. I will clarify

Answer:

Most common type of errors or mistakes could the analyst make:

1.Drawing inferences only basis mean (without removing outliers):

If you believe that any event does not represent a normal outcome, you need to filter it out from your analysis. For example, if a friend of the CEO or a member of Board of Directors invests a big amount through one of your products that should be removed from data. Typically used definition to identify outliers is multiple of standard definition (2x or 3x).

To put numbers on this example, let us say Sales size for your product varies between Rs. 100 – 10,000 and there are 500 such cases such that the average works out at Rs. 5,000. If there is one sale of Rs.5,00,00,000, the average all of a sudden becomes ~ Rs. 1,05,000. If you don’t get this kind of Sale every month, you will end up concluding that average sales size has gone down.

2.Comparing different sets of population, data and clusters:

The example at the start of the article is perfect example of this. This error can happen whenever you do not use non-random way to distribute the population. Whenever you do so, please make sure that populations are similar on all the key parameters.

3. Drawing inferences on thin data:

This happens when you want to bring out insights for every possible segment of data. In doing so, you end up with segments or clusters which have small population and the reads may not be statistically significant (or will contain a lot of noise).

4. Wrong applications of the inferences:

It is human tendency to generalize insights and findings. You extend or apply a learning based on different set of population or circumstances to a different set. Those insights may or may not be relevant. If your credit scoring model was based on a population coming from a particular channel (e.g. Branches), you can not apply it to a different set (e.g. Online). You will end up getting things wrong.

5. Correlation does not mean Causation:

Correlation is a tool every analyst uses on a frequent basis. The biggest caution while using it is that it should not be treated as causation. If two events are happening close to each other, it is not necessary that they are happening because of one another.