Question

I have a very basic question regarding the design of a database. I thought I knew the answer, but after seeing an online tutorial on a triple dropdown menu and the design of the demo database, I’m not so sure anymore.

Ok here’s the setup, there are three tables involved.

First table: tblcountry
Fields: country_id
country

Second table: tblstate
Fields: state_id
country_id
state

Third table: tblcity
Fields: cidy_id
state_id
city

Now my question is: In table three, is it necessary to include the countryid as well? By using SQL you can query which country the city is in….or is it good design to have a reference to the country in the city table??

I hope someone can help 🙂

Answer 1

Normalization starts with data. In what follows, take the word state loosely. I’m talking about data analysis and normalization, not about modeling political hierarchies.

country_name               state_name  city_name
--
United States of America   Alabama     Birmingham
United States of America   California  Atascadero
United States of America   Florida     Key Largo
United States of America   Illinois    Carbondale
United States of America   Kentucky    Winchester
United States of America   Michigan    St. Louis
United States of America   Ohio        Coldwater
United States of America   Montana     Glasgow
United States of America   Louisiana   Jefferson
United States of America   Wisconsin   Jefferson
Bulgaria                   Montana     Montana
Costa Rica                 San José    San José
Argentina                  Entre Rios  San José
Uruguay                    San José    San José de Mayo

It means “City [city_name] is in state [state_name] of country [country_name].” (What a table means is called its predicate.)

This table effectively identifies the “full name” of a city as {country_name, state_name, city_name}. At the conceptual level, it’s not very different from using {last_name, middle_name, first_name} to identify a person. The difference is that, although two people often do share the same full name, two cities don’t.

It’s clear there’s only one candidate key: {country_name, state_name, city_name}. This table is in 5NF. Substituting ID numbers for text can’t change that.

Now, what’s the first thing you think after you’ve normalized a table to 5NF, and that table turns out to be “all key”? Is it, “Ok, now I need to split the key into more tables?”

I hope not.

To increase data integrity, you can project sensible subsets of the columns, and set foreign key references. Projection, used in this way, has absolutely no effect on the normal form of the original table. If it were in 2NF before this kind of projection, it will be in 2NF after this kind of projection. These tables have slightly different predicates from the original. (Duh.)

-- Predicate: "State [state_name] is in country [country_name]."
-- This table is also in 5NF.
create table states (
  country_name varchar(35) not null,
  state_name varchar(15) not null,
  primary key (country_name, state_name)
);

insert into states
select distinct country_name, state_name
from cities;

alter table cities
add constraint cities_fk1
foreign key (country_name, state_name)
  references states (country_name, state_name);

-- Predicate: "[country_name] is a country."
-- This table is also in 5NF.
create table countries (
  country_name varchar(35) primary key
);

insert into countries
select distinct country_name 
from states;

alter table states
add constraint states_fk1
foreign key (country_name)
  references countries (country_name);

To select countries for populating a drop-down list, you can just

select country_name
from countries
order by country_name;

The user having chosen a country, selecting states is simple.

select state_name
from states
where country_name = '?'
order by state_name;

And choosing city names is similar.

select city_name 
from cities
where country_name = '?' 
  and state_name = '?'
order by city_name;