How to create aggregate data

We want to run a survey about how many people like hats.

Open a python shell in interactive mode (run python3) and run the following commands in sequence.

First we’ll import the library and make a new store.

from ocdsmetricsanalysis.library import Store
store = Store("how-to-create-aggregates.sqlite")

We’ll add our own metric.

store.add_metric("HATS", "How many people like hats?", "We ran a survey to find out.")
metric = store.get_metric("HATS")

Here are the results of our survey:

survey_results = [
     {"response":"like", "person_height":"tall"},
     {"response":"like", "person_height":"tall"},
     {"response":"dislike", "person_height":"tall"},
     {"response":"neither like or dislike", "person_height":"tall"},
     {"response":"like", "person_height":"tall"},
     {"response":"neither like or dislike", "person_height":"tall"},
     {"response":"like", "person_height":"short"},
     {"response":"like", "person_height":"short"},
     {"response":"neither like or dislike", "person_height":"short"},
     {"response":"like", "person_height":"short"},
     {"response":"dislike", "person_height":"short"},
     {"response":"neither like or dislike", "person_height":"short"},
     {"response":"like", "person_height":"short"},
     {"response":"dislike", "person_height":"short"},
     {"response":"dislike", "person_height":"short"},
     {"response":"like", "person_height":"short"},
]

We don’t want to publish individual survey responses in our metrics, as there may be anonymisation issues with that.

Instead, we’ll publish aggregates - how many people answered a certain way?

Also, because we have additional data on height, we can break down the answers by height too.

There is a function that will calculate the observations for us automatically.

metric.add_aggregate_observations(
    survey_results,
    "response",
    "answer",
    idx_to_dimensions={"person_height": {"dimension_name": "height"}}
)

Lets go through the parameters:

• survey_results is the array of our survey responses.

• “response” is the key in each survey response we are counting the answers to (normally a survey would have more than one question, so we need to know which one to count!)

• “answer” tells us what dimension in the observations we create we should store the answer in.

• idx_to_dimensions tells us about additional questions in the results that we can use to make more dimensions. In this case we are saying the “person_height” key in each survey response can be used to make a new dimension called “height”.

Let’s quickly list all the observations for all overall survey:

observation_list = metric.get_observation_list()
observation_list.filter_by_dimension_not_set('height')
observations = observation_list.get_data()
for observation in observations:
    print("OBSERVATION id=" + observation.get_id())
    print(observation.get_measure())
    print(observation.get_dimensions())

You should see something like:

OBSERVATION id=000000001
4
{'answer': 'dislike'}
OBSERVATION id=000000002
8
{'answer': 'like'}
OBSERVATION id=000000003
4
{'answer': 'neither like or dislike'}

Let’s also see our results broken down by height:

observation_list = metric.get_observation_list()
observations_grouped = observation_list.get_data_by_dimension('height')
for height, observations in observations_grouped.items():
    print("HEIGHT IS " + height)
    for observation in observations:
        print("OBSERVATION id=" + observation.get_id())
        print(observation.get_measure())
        print(observation.get_dimensions())
    print()

You should see something like:

HEIGHT IS short
OBSERVATION id=000000004
3
{'answer': 'dislike', 'height': 'short'}
OBSERVATION id=000000006
5
{'answer': 'like', 'height': 'short'}
OBSERVATION id=000000008
2
{'answer': 'neither like or dislike', 'height': 'short'}

HEIGHT IS tall
OBSERVATION id=000000005
1
{'answer': 'dislike', 'height': 'tall'}
OBSERVATION id=000000007
3
{'answer': 'like', 'height': 'tall'}
OBSERVATION id=000000009
2
{'answer': 'neither like or dislike', 'height': 'tall'}