When developing such assessment inventories, thorough evaluation is needed, including several dimensions of validity testing. Criterion validity tests the performance of an instrument by comparing it to a gold standard [7]. A rigorous but logistically-challenging method for assessing criterion validity for a home food inventory is to require participants and study staff to independently complete a home food availability instrument, and then compare their responses using the staff report as the "gold standard." Comparisons are evaluated for consistency of results between the two methods using kappa or correlation statistics. In addition, comparisons may be made by examining sensitivity (i.e., proportion of foods in the home assessed as present by the staff that were accurately identified as present in the home by the participant) and specificity (i.e., proportion of foods in the home assessed as absent by the staff that were accurately identified as absent in the home by the participant). Construct validity is suggested when expected relationships are shown between the measure and other variables in a conceptual framework [7]. For example, if an HFI shows high availability of high fat foods in the home, one might expect family members to report a high calorie intake in their diet.

The development of many of the home food availability measures described in the literature include a very limited number of foods, and have not included comprehensive validity testing, particularly criterion and construct validity. Criterion validity is very important if participants in research studies are expected to complete self-report home food inventories on their own while in their homes. Construct validity is also important because an instrument should be associated with expected health outcomes. Only four studies have demonstrated criterion validity of home food inventories [2,8-10], and only two have demonstrated construct validity [2,8], and each has its limitations (as described below).

Crockett and colleagues [8] developed a shelf inventory of 80 foods in 12 categories based on foods targeted in a cancer risk reduction program. Only perishable foods targeted in the intervention program were included on the inventory. They conducted two criterion-related validation studies that compared participant and staff-reported inventories. In the two studies, sensitivity was reported as 0.86 and 0.87, and specificity was reported as 0.92 and 0.90, respectively. In addition, Cohen's kappa showed significant overall agreement (p < .0001) between participant and staff reports in both studies. Furthermore, construct validation was assessed by comparing inventory responses with food frequency questionnaires, with overall agreement of 73.6%

Miller and Edwards [9] assessed the face, content and criterion validity for a 166-item shelf inventory that was based on previous work [8,11] with modifications for fat- and sugar-modified foods that would be relevant to the purchases of individuals with diabetes. Similar to the previous inventory, mostly perishable foods were included. Thirty-one older adults diagnosed with Type 2 diabetes completed the inventory and within 48 hours study staff directly observed foods in the home. Cohen's kappa statistic was 0.87 and sensitivity and specificity were 0.90 and 0.97, respectively. Construct validity was not assessed.

Marsh and colleagues [10] conducted a study with 48 parents that demonstrated criterion validity for a 34-item fruit, juice, and vegetable availability questionnaire. Parents reported the presence of these foods in the home within the last seven days, and staff conducted inventories on the same visit. In addition to perishable foods, frozen, canned and dried fruit/vegetable products were included. Cohen's kappa values ranged from 0.24–0.53 for juices, 0.12–0.76 for fruits, and 0.22–0.66 for vegetables. Overall sensitivity and specificity values for total fruit, juice and vegetables were 36.8 and 39.1, respectively. Construct validity was not assessed.

Lastly, Raynor and colleagues [2] completed a type of criterion validity and construct validity of a household food availability instrument (22 high-fat and 22 low-fat items) with 165 adults. Criterion validity was accomplished by correlating the number of high-fat and low-fat food items reported by two adults in the same household (r = 0.69, p < .001 and r = 0.59, p < .001, respectively). However, gold standard criterion validity comparing participant report to trained staff report was not completed. Construct validity considered correlations between availability of high-fat foods and low-fat foods with fat consumption via a food frequency questionnaire, and showed significant (p < .001) correlations (r = 0.25 and r = -0.33, respectively).

The demonstrated criterion validity of the new home food inventory as shown by high kappa, sensitivity, specificity values and high correlations between participants' and staffs' reports of foods in the home suggests that the instrument could be used effectively for data collection by participants, thus, alleviating the need for staff home visits which are expensive, time-consuming and potentially intrusive. In comparing our criterion validity indices to previous research, kappa, sensitivity, and specificity values appear to be similar to two previous studies reporting these values [8,9], and substantially higher than those reported by Marsh and colleagues [10]. Raynor and colleagues [2] are the only previous investigators to demonstrate significant criterion validity of groupings of high-fat and low-fat foods by showing significant correlations between reports of two adults living in the household. Our study examined several validity indices regarding staff and participant reports of regular fat and reduced fat versions of dairy, added fats, frozen and prepared desserts, and savory snacks and demonstrated substantial validity as well, and extended the previous work by examining subgroups of foods within major categories (e.g., cheese within the dairy category) rather than grouping all low-fat foods together. However, it should be noted that, although the comparison of staff and participant responses is an accepted practice for measuring criterion validity, this testing does not attest to the inventory's capture of all relevant foods, nor does it eliminate the possibility that participants may have altered their responses since the research staff were in their homes.

Although all of the HFI major food groups and many of the food subgroups showed substantial criterion validity, several of the food subgroups did not perform as well and deserve mention. In particular, the reduced-fat frozen dessert, regular fat prepared dessert, and white bread categories had lower than desired criterion validity. Our findings regarding lower validity for prepared desserts is similar to that found in previous research [8]. Several anecdotal observations early in the study indicate that the wide variety of dairy, soy and other frozen desserts available in the marketplace may make it difficult to assess nutrient and fat content. Similar observations were made for prepared desserts. In addition, the proliferation of whole grain white breads and light wheat breads available today confuse participants and staff alike when categorizing bread types.

Our findings regarding construct validity are similar to those reported by Raynor and colleagues [2]. In the present study, all of the correlations between the HFI major food categories and DHQ servings and nutrients were statistically significant in the expected direction. Several correlations between the HFI major food categories and child reported servings and nutrients were significant but attenuated in comparison to those of their parents. Perhaps it should be expected that food availability and intake would be more similar from the same informant (in this case parents) either because he/she purchases foods he/she prefers and eats. Another potential explanation for the poorer construct validity for the youth is that the youth's dietary intake was assessed using 24-hour recalls while the parents completed a food frequency that assessed usual food intake over a longer period of time. Further, our findings that an obesogenic food availability score for the household is significantly and positively associated with energy intake of both parents and adolescents indicate that high fat foods available in the home and captured on the inventory are potentially good starting points for public health messages for healthful eating.

The use of previous instruments and literature associated with energy consumption to determine the selection of foods for the instrument makes the home food inventory useful for many purposes. Previous home food availability measures were developed for specific study objectives such as fruit/vegetable consumption or foods associated with cancer. The broader selection of foods in our inventory increases its utility in nutrition- and obesity-based intervention programs.

There are several limitations that should be noted when interpreting the results of the present study. The present study did not assess test-retest reliability and therefore cannot address consistency of foods available in homes over time. However, consistency of foods may be less of an issue given that Raynor and colleagues [2] conducted two-week test-retest reliability of the absolute number of high-fat and low-fat foods and showed substantial stability. We also did not assess time since last shopping trip which could have influenced the home availability of perishable items or preferred foods which may be consumed more readily [6]. However, our significant correlations between the HFI scores and the DHQ food servings and nutrients indicate that perhaps this potential confounder was not influential. Another potential limitation is that the new home food inventory does not assess quantity; participants either check "yes" if the food is present in the home or "no" to indicate that the food is not present in the home. Accordingly, a household may "score" high on the number of fruits and vegetables even when quantity is limited or low on the number of fruits and vegetables when quantity is high for only a few foods. In addition, the list of foods is not an exhaustive list of all possible foods contributing to the obesity epidemic; however, in selecting foods to be included on the inventory, we balanced the number and types of foods with response burden. Our goal was to create an inventory that was simple and quick to complete. Attempting to collect data on more foods, quantities of foods, or more specifics about foods such as brand, would have impacted the time and complexity of completing the inventory and added to response burden. Moreover, our construct validity testing indicated that the foods measured on the inventory were significantly associated with energy intake from other measures, suggesting that we have captured a significant amount of the variation in the adult diet. Furthermore, the sample used for construct validity (Sample 2) over represented educated, Caucasian adults and findings may not generalize to less educated or minority populations.