• More Blogs
• Observations »
• Bering in Mind »
• Cross-check »
• Extinction Countdown »
• Solar at Home »
• Expeditions »
• Guest Blog »
• Anecdotes from the Archive »

Last spring, British researchers hit on what seemed like a startling finding: Eating lots of cereal before getting pregnant was associated with conceiving a son. Never mind that sex is determined by chromosomes in the father's sperm. The apparent link between gender and diet generated buzz.

But it turns out cereal may not be your lucky charm if you're hoping for a boy.

Today, another group of scientists is disputing that study, charging that its analysis was flawed and its conclusions due to chance. The researchers—from the National Institute of Statistical Sciences, Cornell University and New York Medical College—report their findings in today's Proceedings of the Royal Society B, the same journal that published the cereal study last April.

The U.S. group re-ran data from the large nutrition survey from which the cereal conclusion was drawn, using a different statistical analysis than the University of Exeter scientists. "By our analysis, it looked like everything was random and there was nothing special about cereal," says Stan Young, assistant director for bioinformatics at the National Institute of Statistical Sciences, an independent, nongovernmental think tank in Research Triangle Park, N.C.

The Exeter study asked some 700 British women to answer a 133-item questionnaire about what foods they ate in the weeks prior to and after conceiving, and then months into their pregnancies. Pre-pregnancy cereal breakfasts were associated with birthing boys.

But in any study with that much data, a small percentage of the results will inevitably show statistical significance just by chance, Young says, and that's likely what happened. Still, the findings of such observational studies (which draw conclusions based on what's observed in a group of subjects, versus measuring the effect of a specific intervention on groups in a randomized controlled study) are replicated only 10 to 20 percent of the time, he says.

"We don’t think that if the study was [repeated] that cereal would show up as significant," he says. "Something else would, just by the chance nature of doing these large, complex studies. We think their analysis is wrong and doesn’t support their claim."

Young added, "It's not biologically plausible that the nutrition of the mother has anything to do with the gender of the child."

Fiona Mathews, the lead author on the cereal study, tells us that she ran her data again using Young's methodology.  "We are absolutely confident that our results are statistically robust," Mathews says.

"Sex ratios at birth can be affected by a range of factors including birth order (whether infants are first-born or not), the timing of insemination, and environmental conditions experienced by the mother at conception," she writes in an official response to Young's paper. "The argument of implausibility is therefore erroneous: mothers do, albeit unknowingly, influence the gender of their children.

"We found that eating breakfast cereal, the main breakfast food in the U.K., was associated with having a male infant," she writes. "This may be, as explained our original work, because breakfast cereal is a major contributor to calorie and nutrient intakes, and so it is a 'proxy' marker of high nutrient intakes."

Updated at 1:10 p.m. Jan. 14 with comment from Mathews.

Image © iStockphoto/Jani Bryson

5 Comments

1. 1. seaoftea 06:51 AM 1/14/09

   I can't imagine that this would be the first conclusion drawn even if the data was found to be correct. More likely would be that women may get cravings for specific foods when carrying a child of specific sex.

   Reply | Report Abuse | Link to this

2. 2. ChrisJones 08:54 AM 1/14/09

   Well, duh...

   Reply | Report Abuse | Link to this

3. 3. bliswell 10:33 AM 1/14/09

   "But in any study with that much data, a small percentage of the results will inevitably show statistical significance just by chance,..."
   
   What is the difference between performing one study with a large data set on many questions versus performing many studies asking a small number of questions?
   
   

   Reply | Report Abuse | Link to this

4. 4. Oriel33 01:38 PM 1/14/09

   I think my favorite quote would fit in well here to show a good point about this study. "There are three kinds of lies: lies, damned lies, and statistics" from Benjamin Disraeli.

   Reply | Report Abuse | Link to this

5. 5. axfv in reply to bliswell 03:56 PM 1/15/09

   bliswell: Statistical significance means it's a small percent that the result was obtained by pure chance, usually 5%. I'm going to use die rolling to demonstrate what can go wrong.
   
   Imagine you roll a die 10 times. You hypothesize the die is loaded and 1 is more likely to occur than the rest. In the experiment, you roll 5 1s and since this would only happen about 1.5% of the time had the die been fair (each number equally likely to come up), you thus conclude that the die is loaded in favor of 1 because your results are statistically significant (<5%).
   
   Now imagine you roll the die 10 times, but now you're looking for any loaded pattern. That is, if the die comes with 5 1s, you'll conclude it's loaded because that could only happen 1.5% of the time by chance, if it comes with 5 2s, you'll conclude it's loaded for same reason, same for 3s, 4s, 5s, and 6s. However, if the die is fair, *one of the numbers* will repeat 5 or more times roughly 9.1% of the time. That is, while the chance that each individual number repeats 5 times is only 1.5%, the chance that any of the 6 will repeat is 9% and therefore no longer statistically significant (>5%).
   
   I haven't read either the original study or the rebuttal from Cornell, so I don't know exactly what happened. I also don't think this exact scenario happened with the original study, as I hope they know better. However, it's possible that a more involved version of this occurred.

   Reply | Report Abuse | Link to this