Main Page/Research/Papers/fast food and arterials/IJBNPA/comments

Comments from IJBNPA reviewers, my responses in bold

Revised draft due March 23, 2009

Reviewer 1 Comments for Author

• I have no major concerns with the manuscript. The addition of the arterial road density was a nice addition to the general approach

Minor issues requiring attention:

• Need to clarify the “King County WA” refers to King County in the State of Washington in the US. This should be clarified in the Title and the body of the text

  done

• Page 10 “ These findings are consistent with those studies in the US [3, 11] and the UK [22] that linked fast food restaurant density with area-based measures of low socioeconomic status.” would be more complete if it read “These findings are consistent with those studies in the US [3, 11], the UK [22] and Australia [21] that linked fast food restaurant density with area-based measures of low socioeconomic status.”

  done

• Page 12 (top): Clarify the reference to “MAUP”

  done, added some additional text

Reviewer 2 Comments for Author

Evaluating the contributions of arterial road density to neighborhood fast food density is relatively novel.

Major issues requiring attention:

• The background is not particularly well written, especially the first 3 paragraphs. It is not presented in a logical sequence and thus the main points are obscured. Sentence seem to be misplaced/out of order (e.g., why is the last sentence include in the paragraph on socioeconomic variations in fast food density?)

  We have revised the background; we hope it comes across more clearly.

• Further, the literature review is “off” – there are key studies missing (e.g., a national study on fast food availability by Powell) and some of the citations are not correct. (For the literature review, see the new Larson article in Am J Prev Med 2009 for a review and then see article mentioned within.) For example, the Detroit study examined supermarkets; there is no citation for the Los Angeles study

  [correct; deleted this reference];

• and the cited Powell 2006 and Moore study do not directly measure “access to fresh nutritious foods, including high quality produce.” Some statement seem somewhat contradictory (e.g., fast foods being clustered in more densely populated neighborhoods and in commercially zoned areas – perhaps both are possible but it is not clear how as written).

  References as stated have been corrected.

  FFR require a threshold of clientele to be viable; the clientele may come from workers (in commercial zones) or residents (near areas of sufficient residential density). These are not necessarily at odds.

• The hypotheses are severely underdeveloped. For example, how are area socioeconomic status and arterial road density related? Might road density mediate the effect of area SES (I don’t know)? Neighborhood racial composition is not included in the “hypothesis”; yet, it is included as one of three independent variables.

  We did not directly consider whether area SES and freeway/arterial density would be related. However, it is plausible that real estate prices are lower near highly trafficked roadways, leading to collinearity between FFR density and SES.

• There are a number of problems in the methods section, especially in the description of the measures. (A strength is the use of parcel polygon centroids to calculate distances between residential dwelling units and fast food outlets.)

  The methods section has been reorganized.

• First, based on a list from the public health department, how did you determine whether outlets met your definition of a fast food restaurant? You may know for local and national chain but what about independent restaurants?

  Classification for chains was straightforward; for those food sources that were one-offs, data were reviewed on a record-by-record basis, checking against online data sources including Google, Yahoo, and Yelp.

• Second, the descriptions of your variables require improvement. Follow a format that includes variable, conceptual definition, and operational definition.

  We have clarified the description of variables in the beginning of the Analysis section.

• For example median household income should be identified as your measure of area socioeconomic status (as written in your hypothesis). How was “race aggregated into a dichotomous variable” – you do not state here that you used percent white and not clear how you dichotomized that variable.

  We considered both race and median household income SES variables (now we state "SES (income and race) data at the census tract level..."). Also it was not clear in the original manuscript how we recoded race data. We simply added all nonwhite resident counts per tract to obtain a total number of nonwhite residents. Divided by total residents per tract gave us % nonwhite residents per tract. This was indeed not "dichotomizing" and should be more clear in the revision: "Because of the low proportions of some of the racial subgroups in the county, race was aggregated into a single variable representing the percent of nonwhite residents per tract."

• (Plus, how did you obtain income and race data from TIGER/Line data sets?)

  We stated "Income and race data at the census tract level were obtained from the 2000 US Census SF3 and TIGER/Line data sets," which indicates that income and race data were obtained from the combination of census data (for tabular data) and TIGER/Line (for GIS layer data). We clarify this in the revision, "SES (income and race) data at the census tract level, were obtained from the 2000 US Census SF3 and tract boundaries were obtained from TIGER/Line data sets."

• For the road variable, you should define the different road classes and provide a lay description of these different road classes.

  These are standard FHWA classes; we have clarified that we did not develop these classes. A reference is not added for the FHWA classification system.

• Here you should also indicate this is density per area, rather than population.

  We have clarified that the overlay/summary method generates N of FFR per tract and sum(freeway/arterial length) per tract. From these area or population normalized densities can be developed.

• The outcome variable is not well described. You should clearly state here that you are measuring density both in terms of population and geographic area. Moreover, while perhaps a disciplinary difference, your description of “offsets” and “normalizing” variables is unclear. Are you referring to the same or a different approach?

  "Offsets" are used in generalized linear models, a component of the predictor that is known in advance and requires no parameter to be estimated. Also to decrease possible obfuscation we have removed reference to "normalization" and simply use "area-based" and/or "population-based" density.

• The presentation of the results is also very difficult to follow. Consistency in language is important. Can you use “minor” or “local” consistency?

  We standardized language for street classification "local/minor" and "freeway/arterial".

• Please check how you refer to your density measures – they are called “area-normalized density,” “population-normalized density,” “density,” “density per km2,” “fast food restaurants per capita,” etc.

  • We have attempted to clear this up by removing reference to "normalization" and use "area-based" and/or "population-based" density. In some cases it should be clear that area-based density is used (e.g., "Fast food restaurant density ranged from 0 to over 22 FFRs per km^2"; the reader should know without further instruction that this is an area-based density.
  • Would the reviewer prefer identical terminology throughout the paper? E;g;. "Each additional $1,000 drop in median household incomes was associated with an increase in fast food restaurant density of 5.8% (restaurants per km2) and 3.6% (restaurants per capita).: should be fairly clear that we are referring to area and population-based densities.
  • In one instance we have discussed both area and population based densities and then have made a summary statement "In King County, fast food restaurant density was associated with low incomes, but not necessarily with minority status." which really refers to both types of density.

• You refer to the racial composition measure as “percent white,” “minority status,” “percent nonwhite,” etc.

  We have reviewed the manuscript and standardized to "percent nonwhite."

• (There are also formatting issues in this section which I assume are not the author’s fault.) In some places, I am not sure how the authors reached a conclusion (e.g., page 9, “arterial road density was the best predictor of fast food density” and “arterial road density was only minimally associated” with the other two independent variables).

  We removed the model containing only road density and also removed the statement indicating that freeway/arterial road density was a better predictor of FFR density, and its association with the other 2 variables.

• Points in the discussion are not well-supported. I am not convinced that use of road density measures would change our understanding of the relationship between neighborhood demographics and fast food availability. (My reading of this literature is that findings are mixed.) It may be that low income or minority neighborhood have more fast food restaurants because they have more major roads but this does not change the observation (when found) that they have more. In fact, road density may explain the association.

  An implication exists in the nutrition/obesity/fast food location literature that fast food franchises target the poor. We have shown that in the largest center of population in the 2nd most populous Western US state that, although there is a relationship between fast food restaurant density and SES at the tract level, that built environment layout also plays an important role in the siting of fast food restaurants. Strategies for regulating fast food restaurant locations may benefit by taking this into consideration (e.g., by limiting signage or drive-through access). At the most disaggregate level of available data (dwelling units) we also found that there is a low correlation between dwelling unit value (a proxy for wealth) and proximity to either fast food restaurants or freeway/arterial roadways. We have also shown that for an albeit fairly affluent region, that fast food restaurants are easily physically accessible for nearly all residents, rich or poor.

• Also, the last paragraph is a large leap from your study (and the overall point is obscured, in my opinion).

  Perhaps, but we have shown for this region that fast food is easily accessible to all residents; clearly these restaurants would not exist unless they also had sufficient levels of clientele. If fast food consumption is a major culprit in the obesity epidemic, why then do obesity rates vary so much -- and by SES -- within King County, when fast food is ubiquitous?

  One weakness of this and nearly all other studies of fast food restaurant location and SES or obesity is the lack of measurement of fast food consumption by those with lower SES &/or higher BMI. We will never be able to ascertain causality until we are able to accurately measure at an individual level actual fast food consumption, fast food restaurant access, SES, and BMI. Understanding the complex relationships between environment and behavior require a keen understanding of the physical environment; we believe this study is a step in the right direction of considering the aspects of built environment relevant to the physical configuration of local foodscapes.

Minor issues requiring attention:

• There are a number of typographical and grammatical errors that detract from the manuscript’s quality.

  We have reviewed for typographical and grammatical errors to the best of our ability.

• Street network distance is preferred over Euclidean distance. Please use network distance of justify the use of Euclidean distance.

  Calculation of street network distance between the > 700,000 residential parcel centroids to the 620 fast food restaurants would have been prohibitive in processing time. We have not used distance in any of the statistical models, but have rather used this as ancillary data to show the ubiquity of fast food restaurants throughout the county. Nevertheless, in response to this salient issue we used costdistance modeling to obtain network distances in a reasonable amount of time.

• Why did you convert parcels for properties within multiple residential units to multiple records?

  This was done to calculate mean distance between dwelling units and fast food restaurants. Using only parcels would give the same distance weight to a large apartment building as a single family house. We wanted to develop a measure that reflected household-level, rather than building-level "exposure."

• For the data analysis, why estimate three separate models – why not just present model 3?

  We were interested in the difference between models that used and did not use freeway/arterial density. While models using only demographic variables have shown significant associations between fast food restaurants and SES, these studies generally do not use other -- and potentially structurally important -- built environment variables. We showed that including just one of these variables -- the density of freeway/arterial roadways -- made a significant addition to the model, and substantially attenuated the effect of SES variables. This would not have been possible to show with a single model. We have, however, simplified the modeling into 2 separate model classes: one using only demographic variables, and the other using these demographic variables with the addition of freeway/arterial density (we removed presentation of the model using only freeway/arterial road density and no SES variables).