# Should the units (e.g. currency) be the same in one variable when running regression?

I am doing a regression that involves many countries. One independent variable is revenues of sampled companies in each country. But the data I have use the reference countries' currencies (so I have multiple currencies).

The model requires me to scale the independent variables such as revenues using lagged assets (e.g., revenue(t) in Yen divided by assets(t-1) in Yen).

So the question is, do I need to have a uniform currency say US\$, or won't the outcomes change even if the currencies are different, as long as the currencies used are the same in each company? • When you convert to another currency, the scaled values will not change, unless the exchange rates between the currencies vary over time. Is this what you're trying to get at? – whuber May 31 '17 at 2:04 ## 1 Answer As I understand it, you have a variable that is the average revenue of sampled companies in different countries. The problem with having the different countries using different currencies is that they are no longer comparable at all. The regression model has no knowledge that these cases have been measured on different scales and so would, as a basic example, treat an average revenue of 200,000 Yuan (which is actually only around USD$29, 000) as larger than USD100, 000.

Thus, you will want to convert all of your values into a comparable value such as converting all of them into USD or maybe converting all of them into some measure of spending power (I'm not an economist so I don't know what metrics are used for this but maybe something like how many loaves of bread the revenue would allow you to buy).

Edit: I am not familiar with the concept of lagged assets so I'm not sure whether what you refer to there is converting the company revenues into some comparable metric.

• Doesn't your argument imply, for instance, it would be impossible to use regression to evaluate the association between height and weight in humans, because height and weight are "no longer comparable at all"? Since obviously such a thing is possible and routine, there seems to be something fundamentally wrong with your answer. Or have I misread it somehow? – whuber May 31 '17 at 2:06
• If height and weight are separate variable then it would be fine for them to be on different scales. My interpretation of the question was that cases within a variable were being measured on different scales (e.g. Some people had their height measured in feet, other people had their height measured in cm) – Cam May 31 '17 at 2:21
• @cam, How about the fact that the regression involves a similar currency. Say, in Y = a + b1x1 + b2x2, the y and x's are all in say the currency Yuan for one company. And for the other company, different currency, but same currency for y and x's. Doesn't that make the relationship between the dependent and independent variables still valid? – maroe Jun 1 '17 at 2:37