• Inspect Frequency Table
• Extract Leading Digits
• Inspect Which Values Couldn't be Converted
• Inspect Final Results

Recently, one of our clients used a text field for asking his respondents’ ages. The resulting age variable is in age-in-string.sav, partly shown below.

I hope you realize that this looks nasty:

• age is a string variable so we can't compute its mean, standard deviation or any other statistic;
• we can't readily convert age into a numeric variable because it contains more than just numbers;
• a simple text replacement won't remove all such undesired characters.

For adding injury to insult, the data contain 3,895 cases so doing things manually is not feasible. However, we'll quickly fix things anyway.

Inspect Frequency Table

Let's first see which problematic values we're dealing with anyway. So let's run a basic frequency table with the syntax below.

Result

If we scroll down our table a bit, we'll see some problematic values as shown below.

This table shows us 2 important things:

most values that can be corrected start off with 2 digits;
at least one value is preceded by a leading space.

Let's first remove any leading spaces. We'll simply do so by running compute age = ltrim(age).

Extract Leading Digits

We'll now extract any leading digits from our string variable with the syntax below.

So what we very basically do here is

• we create a new string variable;
• we LOOP through all characters in age;
• we evaluate if each character is a digit: char.index returns 0 if the character can't be found in '0123456789'.
• if the character is a digit (DO IF), we'll add it to the end of our new string variable;
• if the character is not a digit (ELSE), BREAK ends the loop for that particular respondent.

This last condition is needed for values such as “55 and will become 56 on 3/9” We need to make sure that no digits after “55” are added to our new variable. Otherwise, we'll end up with “555639” -an age perhaps only plausible for Fred Flintstone.

Inspect Which Values Couldn't be Converted

Let's now inspect which original age values could not be converted. We'll rerun our frequency distribution but we'll restrict it to respondents whose new age value is still empty.

Result

Surprisingly, a quick scroll down our table shows that we can reasonably convert only a single unconverted age value: “Will become 56 on the 3rd of September:-)”

It is probably safe to infer from this statement that this person was 55 years old at questionnaire completion. We'll set his age to 55 with a simple IF command. We'll then run a quick final check.

Final Frequency Table

As shown below, our minimal corrections resulted in a mere 148 (out of 3,895) unconverted ages. A quick scroll down our table shows that no further conversions are possible.

We'll now convert our new age variable into numeric with ALTER TYPE and inspect the result.

Inspect Final Results

First off, note that our final age variable has N = 148 missing values -just as expected. It is important to check this because ALTER TYPE may result in missing values without throwing any error or warning.

Next, a histogram over our final age values is shown below.

Although the age distribution looks plausible, the x-axis runs up to 120 years. SPSS often applies a 20% margin on both sides so this may indicate an age around 100 years.

Closer inspection shows that somebody reported an age of 99 years. As we think that's not plausible for the current study, we set it as a user missing value.

Done.

Thanks for reading!