Ok smart people! How would you evaluate macro impact on overeating (and possibly satiety)?

EvgeniZyntx

Recently someone wanted to look into their macros and see how it impacted their recent binges.
I wrote a method to do this out of MFP data and will post it below. I'm looking for criticism and input. Does it make sense to you? Who would you do this differently?

General Method: Take the days that are high and look at macro distribution over the prior x days.

First Results: I see a shift where I tend to eat less fat the few days prior to a "large" day. (Note, I don't have a binge issue, so these are just days with a few hundred cals over.)

And this is purely correlation - I'm not saying that at this time that it actually impacts my satiety. I'm not tracking "satiety" in my dairy.



Detailed Method:

Select only days with calorie data for the prior x days.
For the days with high calories, take average of protein, fat and carb macros for the prior x days.
Normalise these averages against all days with calories.
Look at distribution of these days - do they tend to be the same as general eating or is there a shift towards higher or lower days?

Thoughts? Comments?

MelodyandBarbells

That's pretty neat. Does each person's diary in this case equals N, or an entirely separate experiment? I wonder because "high calorie day", for example would have to be determined on an individual basis, and would also likely be randomly selected. Or not. I suppose you Could use a formula like 1.XX TDEE, who knows.

By analyzing this statistically, you should be able to show if there's significant impact of macro distribution in days preceding an overage day. This should also indicate if there's no link, that is, other factors such as a holiday gathering, emotional eating, lack of planning, I'm in Momma's house so I better eat ALL THE THINGS, ETC have come into play. However, macros seem like an easy one to evaluate given our data

Oh also the individual would of course need to have been tracking macros properly. Even now, I mostly double check macros on individual food items when one entry throws my pie chart WAY out of whack. This has happened a total of two times

lemurcat12

Is there a way to input the number of days before the high cal day that you want to look at so that you can try different options? For example, first look at day before then the 2 days before?

Also, would the user input what a high cal day is? Days above 2000, say? Or would it take the average calories for a period and look at days that are more than some percentage above the average?

I don't think it would work for me, though, since I have lots of days that are planned highs -- either an evening out or a big exercise day. But that aside, it seems like a really great tool.

EvgeniZyntx

JaneiR36 wrote: »

That's pretty neat. Does each person's diary in this case equals N, or an entirely separate experiment? I wonder because "high calorie day", for example would have to be determined on an individual basis, and would also likely be randomly selected. Or not. I suppose you Could use a formula like 1.XX TDEE, who knows.

The idea is a personal analysis - not research across the database - 1) because I do not want to look at other people's data without permissions or manage passwords - one just downloads the excel sheet and runs it off their own data 2) it is pretty easy to see high days - just take target cals and identify days that are above that or use a limit that is manually set. Currently, I'm working with a manual set limit because that allows me to play with it.

By analyzing this statistically, you should be able to show if there's significant impact of macro distribution in days preceding an overage day. This should also indicate if there's no link, that is, other factors such as a holiday gathering, emotional eating, lack of planning, I'm in Momma's house so I better eat ALL THE THINGS, ETC have come into play. However, macros seem like an easy one to evaluate given our data

Since I can't exclude causes, this is not a causal analysis but a correlative stats. If you can't eat to a target, the solution might be to move out of Momma's but this will tell you if you are eating more carbs or fat on the days before.

You can see that my protein distribution is bi-modal. That is because I also have high days were I purposefully eat above cals to reach my protein goals. The analysis won't give the context.

Oh also the individual would of course need to have been tracking macros properly. Even now, I mostly double check macros on individual food items when one entry throws my pie chart WAY out of whack. This has happened a total of two times

Yes, bad data makes bad analysis.

EvgeniZyntx

lemurcat12 wrote: »

Is there a way to input the number of days before the high cal day that you want to look at so that you can try different options? For example, first look at day before then the 2 days before?

Also, would the user input what a high cal day is? Days above 2000, say? Or would it take the average calories for a period and look at days that are more than some percentage above the average?

I don't think it would work for me, though, since I have lots of days that are planned highs -- either an evening out or a big exercise day. But that aside, it seems like a really great tool.

Absolutely, the idea is to let the user set the number of consecutive days to consider. It recalculates the curve as one changes the number of day to consider.

yarwell

are the binges logged accurately ? If so some form of multivariable regression might be helpful.

MelodyandBarbells

Ok, I'll be th dumb one. What are the X and Y axes on the graph? I see Y goes up to 16, but can't really tell what unit that is. Being normalized probably means there's no unit, but perhaps there's some context I'm missing

EvgeniZyntx

yarwell wrote: »

are the binges logged accurately ? If so some form of multivariable regression might be helpful.

Let's assume they are. I'll look into that idea. A bit more complicated.

EvgeniZyntx

JaneiR36 wrote: »

Ok, I'll be th dumb one. What are the X and Y axes on the graph? I see Y goes up to 16, but can't really tell what unit that is. Being normalized probably means there's no unit, but perhaps there's some context I'm missing

It's a normalised histogram

y is a count - number of days,
x is the average of days prior to a high days / the average of all days. It is unit-less.

So think of it this way - your looking at a sample of days (days before the high cal day). If they are similar to the distribution of all days then the curve should look like a bell curve with a median of 1. If the high days have a tendency to be driven by, say, low fat days prior, then we should see the curve shifted to the left. Like in that sample curve.

This is a good question. I need to be clear on it.

Let's say you cut a bunch of strings and you are trying to make them 5 inches long. Some will be little shorter, others a little longer. If you take the length of each, divide by the average, then count all the strings that are between .9 and 1.0 or 1.0 and 1.1, etc you can build a histogram, that is the count for each value (y axis).

ClosetBayesian

yarwell wrote: »

are the binges logged accurately ? If so some form of multivariable regression might be helpful.

Or a time lag analysis in Observation Oriented Modeling.

snikkins

My thoughts aren't 100% clear on this so ask for clarification if I am not making sense.

Are you writing this to pick out certain high calorie days and then look at the day(s) preceeding it? Or will the user select the day that a binge occurs and analysis will happen on the previous day(s)? If it's the latter, then I think there won't be an issue determining which day(s) are being analyzed.

EvgeniZyntx

ClosetBayesian wrote: »

yarwell wrote: »

are the binges logged accurately ? If so some form of multivariable regression might be helpful.

Or a time lag analysis in Observation Oriented Modeling.

Let's keep it real, this runs off Excel.

EvgeniZyntx

snikkins wrote: »

My thoughts aren't 100% clear on this so ask for clarification if I am not making sense.

Are you writing this to pick out certain high calorie days and then look at the day(s) preceeding it? Or will the user select the day that a binge occurs and analysis will happen on the previous day(s)? If it's the latter, then I think there won't be an issue determining which day(s) are being analyzed.

Currently written to review all days above a threshold - but in such a manner that you can select one or more specific days.

So the answer is "yes" to both questions.

snikkins

Fascinating. I don't binge, but I think it'd be neat to check out when you're done.

For the first scenario, are you going with a certain threshold above the daily goal?

EvgeniZyntx

snikkins wrote: »

Fascinating. I don't binge, but I think it'd be neat to check out when you're done.

For the first scenario, are you going with a certain threshold above the daily goal?

Two thresholds - a lower one (say 400) anything under is considered a day that wasn't fully logged. A higher one - user set, above daily goal.

The current version of the worksheet is here: https://www.dropbox.com/s/10mhbzig1v4v7h2/mfp6v5.xlsm?dl=0
(to use your own data, go to the "Start" worksheet, press "Reset" and then "Go")

snikkins

Thanks!

dubird

This would be something interesting to see in a larger group sample. Self-reporting data does have it's flaws, but it would be interesting to see the general direction it takes.

EvgeniZyntx

dubird wrote: »

This would be something interesting to see in a larger group sample. Self-reporting data does have it's flaws, but it would be interesting to see the general direction it takes.

Possibly but I also think that would require some cluster classifications - we have people losing, others maintaining, bulking, active, sedentary, athletes, people that binge, people that don't... making inferences from random user data is going to be very difficult. However, if we had two groups - bingers vs non-bingers it might give some interesting results.

senecarr

EvgeniZyntx wrote: »

dubird wrote: »

This would be something interesting to see in a larger group sample. Self-reporting data does have it's flaws, but it would be interesting to see the general direction it takes.

Possibly but I also think that would require some cluster classifications - we have people losing, others maintaining, bulking, active, sedentary, athletes, people that binge, people that don't... making inferences from random user data is going to be very difficult. However, if we had two groups - bingers vs non-bingers it might give some interesting results.

I'd find it interesting if exercise calories were an option for comparison as well.
I believe there is some research that depending on a gene or two some people have different appetite responses to exercise. My own personal feeling lately is that my appetite as a function is almost a normal curve. I could be okay eating nothing all day if I was laying in bed, if I was sitting at work all day and sitting at around 3K steps I'd like to eat 3K kcal of food, but if I'm actually moving 20K+ steps and/ burning an estimate of over 2.5K+, I'd mentally feel okay eating under 1K kcal of food and have to ignore my un-appetite.

Sued0nim

I think you need overlays of liquid intake, exercise - cardio/ resistance and hormonal shift (specifically in women)

robertw486

EvgeniZyntx wrote: »

Recently someone wanted to look into their macros and see how it impacted their recent binges.
I wrote a method to do this out of MFP data and will post it below. I'm looking for criticism and input. Does it make sense to you? Who would you do this differently?

General Method: Take the days that are high and look at macro distribution over the prior x days.

First Results: I see a shift where I tend to eat less fat the few days prior to a "large" day. (Note, I don't have a binge issue, so these are just days with a few hundred cals over.)

And this is purely correlation - I'm not saying that at this time that it actually impacts my satiety. I'm not tracking "satiety" in my dairy.



Detailed Method:

Select only days with calorie data for the prior x days.
For the days with high calories, take average of protein, fat and carb macros for the prior x days.
Normalise these averages against all days with calories.
Look at distribution of these days - do they tend to be the same as general eating or is there a shift towards higher or lower days?

Thoughts? Comments?

Seems to make sense to me, but I do think looking at total calorie intake as well as exercise loads might have impact as well. My heaviest eating days tend to follow a period of hard exercise after a trend of lighter days, so really I am just making up for prior excess deficit.