Common Mistakes Every Analyst is Bound to Make

By Urbi Ghosh, 3EA
Common Mistakes Every Analyst is Bound to Make

Many a times there are circumstances when individuals who deliberately or non-purposefully end up making incorrect interpretations.

The reason for this post is to draw out a portion of the basic errors analysts make and how to evade them:

1. Drawing Inferences without Eliminating Outliers

If you are certain that any event does not characterize a normal outcome, you need to filter it out from your analysis. For instance, if a companion of the President or the CEO or a member of top managerial staff contributes a major sum through one of your commodities, that ought to be expelled from the data.

Following measures will ensure that you are not making any errors in interpreting inferences by looking at mean:
(i)- Check the distribution and screen out any outliers from your exploration
(ii)- Notice the skew of (mean/median) to identify the amount of skew coming from bigger or smaller values. Higher or lower the value (compared to 1), higher is the skew. If the skews are high, consider both mean and median before making any deductions.

2. Comparing Distinct Sets of Population, Segment or Cluster
You can face this error if you do not distribute the population in a non- random way. Whenever you do so, please ensure that populations are alike on all the crucial parameters. It is best to refrain from associating non-random distributed population. In case, you require to do so due to logistic & resource constraints, compare pre vs. post before you arrive at any assumption.

3. Drawing Inferences on Thin Data
This happens when you need to bring out insights for each conceivable portion of data. By doing this, you end up with fragments or clusters which have little population and the peruses may not be statistically significant.
To avoid such blunders, confidence intervals should always be plotted to the values to be calculated or extrapolated. This will give an idea whether the extrapolation is precise or erroneous.

4. Wrong Applications of the Inferences
It is human propensity to sum up insights and outcomes. You widen or apply a learning in light of various set of population or conditions to a distinct set. Those insights could conceivably be significant. In case, your credit scoring model depended on a population originating from a specific channel (e.g. Branches), you cannot have any significant bearing it to a distinct set (e.g. Online). You will wind up misinterpreting things.
To avoid such errors, continuously check whether you are outspreading your models to a population which it has not seen in past. If you know about any progressions, keep on monitoring the population on key factors (e.g. for credit scoring model, you have to screen age, wage, unemployment rate, credit document match % and so on).

5. Correlation doesn't imply Causation
Correlation is a statistical tool every analyst uses frequently. The principal caution when using it is that it should not be considered as causation. If two events are occurring close to each other, it is not obligatory that they are occurring because of one another. For example, if my age is increasing with time and my brother's age is also increasing with time, it does not mean my age is driving his.

These were the most common errors that analysts tend to make. With practice comes perfection. So, the only advice is to practice more with diverse datasets on a regular basis.

#ReadyBusinessPlan #Ask3EA #LearnAt3EA #3EA #BusinessPlan #CapacityEnhancement #CapacityBuilding #Capacity #Assessment #Global #BusinessAnalyst

Article by: Urbi Ghosh, 3EA