Query two tables joined over a third table with the same foreign key? - sql

Question

I have a Postgres database schema groceries.
There are two tables purchases 19 and 20 connected over a third one categories.
I can join every table alone with categories without problem.
For calculating the year change I need 19 and 20 together.
It seems the problem is that the third table categories has got only one foreign key for both tables. Thus it return every time a col with zeros because there is no match for one table. Maybe I am wrong.
Any suggestions to query the tables?
More info below.
The groceries database has a subset dairies: 'whole milk','yogurt', 'domestic eggs'.
There are no clear primary keys.
I share the database file with this link:
https://drive.google.com/drive/folders/1BBXr-il7rmDkHAukETUle_ZYcDC7t44v?usp=sharing
I want to answer:
For each month of 2020, what was the percentage increase or decrease in total monthly dairy purchases compared to the same month in 2019 (i.e., the year_change)?
How can I do this?
I have tried different queries along this line:
SELECT
a.month,
COUNT(a.purchaseid) as sales_2020,
COUNT(b.purchase_id) as sales_2019,
ROUND(((CAST(COUNT(purchaseid) as decimal) /
(SELECT COUNT(purchaseid)FROM purchases_2020)) *100),2)
as market_share,
(COUNT(a.purchaseid) - COUNT(b.purchase_id) ) as year_change
FROM purchases_2020 as a
Left Outer Join categories as cat ON a.purchaseid = cat.purchase_id
Left Outer Join purchases_2019 as b ON cat.purchase_id = b.purchase_id
WHERE cat.category in ('whole milk','yogurt', 'domestic eggs')
GROUP BY a.month
ORDER BY a.month
;
It gives me either no result or the result above with an empty sales_2019 column.
The expected result is a table
with the monthly dairy sales for 2020, the montly market share of dairies of all products in 2020, and the monthly year change between 2019 and 2020 in percentage.
How can I calculate the year change?
Thanks for your help.

Answer 1

%%sql
postgresql:///groceries
with p2019Sales as (
select
month,
count(p.purchase_id) as total_sales
from purchases_2019 p
left join categories c
using (purchase_id)
where c.category in ('whole milk', 'yogurt' ,'domestic eggs')
group by month
order by month
),
mkS as (
select
cast(extract(month from fulldate::date)as int) as month,
count(*) as total_share
from purchases_2020
group by month
order by month
),
p2020Sales as (
select
cast(extract(month from fulldate::date)as int) as month,
count(p.purchaseid) as total_sales,
round(count(p.purchaseid)*100::numeric/ m.total_share,2) as market_share,
sum(count(*)) over() as tos
from purchases_2020 p
left join categories c
on p.purchaseid = c.purchase_id
left join mks m
on cast(extract(month from p.fulldate::date)as int) = m.month
where c.category in ('whole milk', 'yogurt' ,'domestic eggs')
group by 1,m.total_share
order by 1,m.total_share
),
finalSale as (
select
month,
p2.total_sales,
p2.market_share,
round((p2.total_sales - p1.total_sales)*100::numeric/p1.total_sales,2) as year_change
from p2019Sales p1
inner join p2020Sales p2
using(month)
)
select *
from finalSale

Answer 2

The answer of user18262778 is excellent.
but as Jeremy Caney is stating:
" add additional details that will help others understand how this addresses the question asked."
I deliver some details.
My goal:
get the output I want in one query
My problem:
The query is long and complicated.
There are several approaches to the problem:
joins
subqueries
All are prone to circular dependencies.
The subqueries and joins produce results,but discard data necessary to move on further towards the final result
The solution:
The with statement allows to compute the aggregation and reference this by name within the query.
If you know it is the WITH statement, then there is of course a lot of info on the web. The description below summarises exactly the benefits of the given solution in general.
"In PostgreSQL, the WITH query provides a way to write auxiliary statements for use in a larger query. It helps in breaking down complicated and large queries into simpler forms, which are easily readable. These statements often referred to as Common Table Expressions or CTEs, can be thought of as defining temporary tables that exist just for one query.
The WITH query being CTE query, is particularly useful when subquery is executed multiple times. It is equally helpful in place of temporary tables. It computes the aggregation once and allows us to reference it by its name (may be multiple times) in the queries.
The WITH clause must be defined before it is used in the query."
PostgreSQL - WITH Clause