Causal influence of dietitians on obesity: Part 1

Posted by Brandon Schaufele in Food, Obesity, Policy on Mar 24th, 2010 | 0 comments

This is the post that I promised here. 

My discussion of dietitians and obesity contains two parts.  Part 1 concentrates on econometric results.  Part 2 will examine the policy implications in greater detail (this will come next week).

Obesity is a problem in Canada.  Estimates on the direct and indirect costs are generally in the range of $5 billion per year but may be much greater.  Due to the size of this economic burden, many anti-obesity policies have been suggested.  These include advertising restrictions, fat taxes, junk food bans, regulation, etc.  The problem with many of these is that they involve a large change in the current food and health landscape.  Introducing a fat tax, for example, would be a headline grabbing policy.  Incremental proposals are often easier to implement and are more realistic.

This analysis examines the causal impact of the number of dieticians on the obesity rate for 25 Canadian cities.  The data are from 2001 (see here and here for background).

Why focus on dieticians?  There are several reasons.  I’ll discuss two.

First, dieticians are leaders in advancing health through food and nutrition.  They are professionals trained to provide health and diet guidance.  Moreover, when it comes to obesity, I’d hypothesize that dietitians focus on two types of patients: high-risk and marginal.  High risk patients are those that impose the largest economic burden on society.  Helping these individuals manage their diet, health and weight should produce the largest economic gains.  Marginal individuals are on the cusp of obesity.  With a little assistance, they are able return to a healthier weight.  This group imposes a smaller cost on society, but has a larger influence on the obesity statistics.  If we care about the tangible outcomes of policies, these two groups are most relevant.

It is the second reason that captured my interest.  It is possible to influence the number of dieticians with small policy changes.  Policy can’t explicitly alter people’s eating or exercising habits and introducing food regulations or restrictions too often is an all-or-nothing proposition.  With dieticians, it’s different.  Via funding reallocations, it is possible to incrementally increase or decrease the number of professionals in a given region.  This in turn could influence health outcomes, including the obesity rate, for that health district.

But something is missing.  Before I discuss the costs and benefits of hiring more or fewer dietitians, a key question must be addressed: do dietitians actually have an influence on the obesity rate of a city?  And what is the direction and magnitude of the effect?  Well, let’s start with some raw data.

The downward sloping trendline tells a convincing story.  It appears that an increase in the number of dieticians does lead to a decrease in a city’s obesity rate.  This is a good first step.  However, we know that correlation does not imply causation.  Reverse causality and omitted variable problems are likely. 

Okay.  So the figure doesn’t demonstrate causation.  Does this mean we’re stuck?  Can we only say that there is a correlative relationship between dietitians and obesity?  Well, for today at least, there may be a solution: Econometrics to the rescue.  (I can’t tell you how many years I’ve waited to write that sentence.)  Applied economists have developed a range of tools to establish causality using observational data.  And what luck, it just so happens that I am an applied economist.

In order to determine causality, I use a method known as instrumental variables (IV for short).  The theory of IV is well-established, still the method can be challenging to execute in practice. 

(Note this is a wonkish paragraph that is not necessary to understand the main results.)  With IV, it is important to consider which factors drive the error term and endogenous variables.  There are several items that I could discuss, but one point is important.  The regression models include either provincial or regional dummy variables as controls.  This means that the error term is picking up city-specific factors.  This is good news.  Here’s why.  I use several instrumental variables.  The most relevant is the number of students who graduated from a four-year nutrition or dietitian-focused university program within the entire province.  This provincial level variable is likely correlated with the number of dietitians in particular cities within that province.  More importantly, it is probably not correlated with the city-specific error terms.  Graduates from these programs need at least four years of schooling.  Any city-specific factors that motivated these students to initially enrol hopefully disappear before graduation.

Here are some econometric results:

OLS*

Obesity Rate = 0.227 – 0.013*Log(Dietitians) + Control Variables

90% Confidence Interval on Dietitians Variable: (-0.020, -0.006)

IV*

Obesity Rate = 0.316 – 0.032*Log(Dietitians) + Control Variables

90% Confidence Interval on Dietitians Variable: (-0.053, -0.011)

Here’s the key message: A 1% increase in the number of dietitians in a city causes the obesity rate to decrease by 3.2%.

How do dietitians influence the obesity?  This result is probably picking up both direct and indirect effects. 

First, there is the direct effect.  Dietitians deal with issues such as diet and health.  More dietitians mean greater access to food and health advice.  Obesity is largely about food consumption.  It’s pretty simple.

Of course, the total number of registered dietitians in Canada is low, so the indirect effect may be more relevant.  What do I mean when I say indirect effect?  A health authority that hires more dietitians is signalling that it is focused on the influences of nutrition and diet on health.  The number of dietitians is simply a single manifestation of this philosophy.  The health authority may engage in a range of diet-related health promotions beyond direct patient care.  It is this indirect effect – i.e., concentrating on diet’s effect on health in general – that is relevant.

The results from the IV model are convincing.  Still, before we run to our politicians, we should do a few checks.  The data are from a single year, 2001.  It is important to ensure that something odd didn’t happen that year.  Below is a scatter plot of provincial obesity rates against the total (log) number of registered dietitians in a province for 2005.  The trendline indicates that the pattern is similar to the econometric results.  This adds significant credibility to the finding that simply having more dietitians in a city may reduce the obesity rate.  (Hint, hint: Sudbury.)

I would appreciate comments on this post.  I am particularly interested in hearing from registered dietitians and food professionals.  Generally, on technical posts, such as this one, people contact me via the contact us button.  If you have a comment that you don’t want publicized, feel free to use it.  Also, if I’ve made an error, please let me know.

* These models are well-behaved.  The dietitian variables are statistically significant at a 5% level for both models.  The F-stat for the instruments in the IV model is less than 10, but statistically significant (F=6.60, p-value=0.02).  This means that these instruments are relatively good.  Finally, the model is over-identified, so it is possible to test for correlation between the instruments and error term – no problems there either.