This case study demeontres my ability to perform common data tasks such as cleaning , KPI contruction , descriptive analysis and and data modeling
Case study - Analyst
Step 2: Calculating KPIs* One of our main KPIs within the Schools Vertical is “Percent Pupils Present”. The “layman’s definition” of this KPI is “The percentage of pupils who were present, out of all pupils - across all days in the term to date ”. In other words, the percentage of pupils who were present (for each pupil in the “Pupil attendance” file, this is displayed in the “present_records” variable), out of pupils who had attendance records (the “attendance_records” variable in the same file). ● The first task is to translate this KPI into the data. We will calculate this KPI in two different ways. First, calculate this KPI for all pupils at once. What is the network-level average Percent Pupils Present (use two decimal points)? ● Now, I calculated this percentage for each school, and created an average at the school-level. ● How does the interpretation of the KPI change between the two approaches? Does it matter in this case? When would it matter, (i.e., when would one be more appropriate than the other?) 2-4 sentences max.
Sub report from analysis 1)When I used the first approach to calculating the “Percent Pupils Present” I found that the network-average Percent Pupils Present is 76 % when rounded to two decimals . This could be interpreted as the fact that bridge Kenya program recorded across all the pupils an estimation of the Percent Pupils Present as 76% .
- Afterwards I calculated the percentage for each school and averaged it at the school level. The average Percent Pupils Present now is still 76%. 3)The interpretation of the KPI between the two approaches is that for the second approach it can be inferred that this KPI is an estimation of Percent Pupils Present across schools whereas the first approach gives us the Percent Pupil present across pupils. 4)I don’t think it matters because the two approaches give us the same estimation.
Step 3: Descriptives analysis* Let’s dig into the reading fluency scores in your current data set. These came from the “Pupil scores” data, I used the data set I created in Step 1 above to answer these questions.
Sub-report from analysis Grades Average Fluency score Grade 1 53.16521 Grade2 104.43394 Grade 3 127.02271 Grade 4 145.47334 Grade 5 154.75989
This table above shows the average fluency score across all the 5 grades in the data
From the the output we see that The Kirinvaga region has the lowest average fluency score with an average fluency score equals to 60.31 . The Machkos region has the highest average fluency score with an average fluency score equals to 157.76
Grades Share of pupils that reads at 10 or lower Grades 1 0.110 Grade 2 0.034 Grade 3 0.022 Grade 4 0.0166 Grade 5 0.0091
This table above shows the share of pupils that reads at 10 or lower across all he grades
This bar chart shows the share of pupils that reads at 10 or lower across all the 5 grades in the data We see from this bar chart that the share of pupils that reads at 10 or lower goes down from grade 1 to grade 5 What we see here is quite normal since pupils tend to get better at reading as they progress through grades.
I found that the school with the highest share of pupils that read at 10 and lower is the school with the identification number( 223941 in the data set). With roughly 34% of its grade 3 kids who reads at 10 or lower.
Step 4: Impact evaluation During this term, we rolled out an intensive after-school tutoring program in 55 schools. The selection to be a part of the 55 schools was randomly assigned - in other words, these schools were part of a randomized controlled trial (RCT). The “School Information” data set has a binary variable for whether each school was part of the program or not. ● Our Chief Academic Officer would like to know whether this program had any effects on test scores in math, Kiswahili, fluency, and/or student attendance. Please conduct any calculations you see fit to answer his questions. ● After conducting the impact evaluation, we have heard anecdotally that teachers in schools that received tutoring felt more motivated and were completing their lessons at a faster pace. Hence, we could worry that the effects that we see are not (solely) due to the tutoring program, but also due to the higher lesson completion rate. Does this hypothesis hold up in the data?
*Findings
We found that pupils in the schools where teachers were trained tend to read at 33 points more than pupils in the schools that did not take part in the tutoring program.. This can allow us to state that the tutoring program has a significant impact on pupils’ reading abilities.
We found that kids in the schools where the teacher participated in the tutoring program tend to have on average 0.02 more scores than the ones in schools where teachers did not receive the training. This result allows us to conclude that the tutoring Program has a significant impact on pupils’ math scores
We found that pupils in the schools where teachers received the tutoring program tend to score 0.11 in Kiswahili more on average than the ones in the schools where the teachers did not participate in the program. Thus it can be inferred there is evidence that the tutoring program has a significant impact on Kiswahili scores of the pupils.
We are now interested in knowing whether the tutoring program has any significant impact on lesson completion rates
we see that there is no any significant correlation between the program and the lesson completion rates Hence we can conclude that this motivation of the teachers that came from the anecdote may not be caused by the fact that they received the training .
Hypothesis verification From the Tables 5 and 6 we see that when we control for the lesson completion rates variable we do not see any significant correlation between lesson completion rates and test scores ( Kiswahili and reading scores) . We can then inform the Academic officer that the tutoring program could have a causal impact on Kiswahili and reading scores whereas this conclusion may not hold for the score in math since we see a significant correlation between lesson completion rates and math scores (which comes from Table 7 below ).