The purpose of the questionnaire was to assess maternal infant milk feeding practices, mothers' decision making regarding how much and how often to feed their babies, their attitudes to infant feeding and growth, and beliefs about following infant feeding recommendations. The 57-item (19 questions), self-administered questionnaire was developed after conducting literature reviews[^18-21] and a qualitative study using a flexible semi-structured interview schedule involving 38 parents (n = 35 mothers). Three main themes emerged on parents' decision-making regarding volume and frequency of formula-milk feeds; i) baby's appetite (if the baby finished the bottle, or cried between feeds, more was added to the next feed), ii) instructions on formula milk tins/packets (if the baby did not take what it said on the tin, they were offered a feed again after a short gap) iii) baby's growth (as baby's weight increased, feeds were increased). Parents got information on bottle-feeding from various sources -family, friends, other mothers, 'parent support groups', books, magazines, the internet, formula-milk packets, supermarket shelves, health visitors and midwives. The main barriers to reducing formula-milk feeds were concerns that the baby would cry, be hungry, wake up at night and demand more frequent feeds.

An iterative process was used and numerous revisions were made in response to input from groups of researchers and mothers. Content validity was assessed by extensive pilot testing with mothers (n = 60) participating in a birth cohort study (The Cambridge Baby Growth Study)[²²]. Questions covered four domains: 1) type of milk feeding, decision making, and sources of advice, 2) frequency and quantity of feeds, 3) attitudes to infant feeding and growth, and 4) theory-based beliefs about following recommendations to reduce formula-milk feed quantities.

Thirty one mothers were recruited from an ongoing birth cohort study[²²]. To include a more diverse range of participants, we also recruited nine exclusively formula-milk feeding mothers from a focus group study conducted to inform intervention development. To assess test-retest reliability, following receipt of a completed postal questionnaire, the same questionnaire was posted and mothers were asked to complete it a second time (median time interval between completion of the two questionnaires was 6 days (range 2-16 days).

To validate the questionnaire, no 'gold-standard' existed, therefore semi-structured interviews were used to assess criterion validity as previously reported for the validation of a questionnaire covering correlates of children's physical activity[³¹]. We developed an open-ended, semi-structured interview schedule in which participants could make general comments on the questionnaire, clarify the meaning of the questions, and expand on their answers, giving a richer response (see Additional file 1 for interview schedule). After the first questionnaire was returned, an appointment was made for a face-to-face or telephone interview with the mothers. At the start of the interview, it was confirmed that mothers had completed the second questionnaire and they were asked for general comments on the questionnaire. The median time interval between completion of the first questionnaire and interview was 8 days (range 3-16 days). Interviews were conducted by RL or JL, were tape-recorded and were transcribed by an independent company. The interview data were used as the criterion measure against which to compare the participant's first questionnaire responses. For each participant, two researchers (RL and JL) used the transcripts to fill in a blank questionnaire. This was done independently and blind to the participant's questionnaire responses. For the two questions (questions 16 and 19 with a total of 19 items) which were scored on a five-point Likert type scale, the transcripts were used to mark a collapsed three-point scale 'agree, neutral, disagree'. Once all of the transcripts were coded, the two researcher-completed questionnaires (for each participant) were compared, and in the event of disagreement, consensus was reached by discussion. This final 'agreed' (by both researchers) questionnaire was used as the 'criterion' for comparison against the participant's first questionnaire responses to assess criterion validity.

Approvals for the study were obtained from the local Research Ethics Committee and research governance committees of the local hospital and Primary Care Trust. Demographic details and written informed consent were obtained from all participants.

All analyses were performed using Stata version 10.1 (Stata Corp LP, College Station, Texas). The test-retest reliability (between 1^st and 2^nd questionnaires) and criterion validity (between 1^st questionnaire and interview coded response) were assessed for each item by calculating percentage exact agreement and chance-corrected agreement (Cohen's kappa). A percentage agreement ≤ 66% was used to indicate fair agreement [³²,³³] and kappa values were categorized as: poor (< 0.0), slight (0.00-0.20), fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1.0) agreements[³⁴]. Weighted kappas were calculated for ordinal categorical variables and Spearman's correlation coefficients for continuous variables.

To assess internal consistency we calculated Cronbach's alpha for items measuring theory-based constructs- self-efficacy, outcome-expectancies and intentions. Negatively worded items were re-coded. A summary score for each construct was calculated by summing the individual item scores and dividing by the number of items in the construct (4 for self-efficacy, 5 for outcome-expectancies and 2 for intentions). The correlations between a summary score for these constructs were calculated.

The inter-rater agreement between the two researchers coding the interviews was high for 56/57 items (agreement above 80%, kappa above 0.6; data not shown). Overall, for test-retest reliability, percentage agreement was good (above 66%) for 89% (51/57) of the items. For the remaining 6 items (percentage agreement between 47% and 66%), Cohen's kappa's were moderate or substantial (0.41-0.68). Overall, for validity, percentage agreement was above 66% for 68% (39/57) of the items. Two items had insufficient data to estimate percentage agreement and of the remaining 16 items, only five had kappa values less than fair (< 0.20).

All 21 items in this domain had percentage agreement over 73% for test-retest reliability and 71% of the items had percentage agreement over 66% for validity. Of the six items that had less than 66% agreement for validity, four related to what advice was followed and two related to how feeds were prepared.

Regarding decisions on the frequency of feeds, 50% used a combination of demand and routine feeding. When deciding on the quantity of feeds, 93% were led by the baby's appetite. Sixty-three percent of all mothers (25/40) prepared formula-milk from powder or concentrate and of these 88% (22/25) reported following the advice on the packaging about feed preparation. However, 44% (11/25) of mothers who prepared formula-milk from powder reported that they tightly packed the scoops.

Although percentage agreement between the two questionnaires was below 70% for six items (test-retest reliability), all 16 items in this domain had moderate to almost perfect agreement based on kappa values (above 0.41), and 80% (13/16) had kappa values above 0.41 for validity. The three items that had fair validity (Kappa 0.2-0.4) were continuous variables: number of breastfeeds, number of formula-milk feeds, duration of breastfeeds, and had correlation coefficients of 0.9, 0.6 and 0.9 respectively. It is unclear whether the questions are not valid or whether feeding practices changed over the duration between the first questionnaire and interview.

The median daily number of breastfeeds was six and the median duration of each feed was ten minutes. The median number of formula-milk feeds was four and median duration was 15 minutes. Eighty percent of mothers had introduced solid foods and 10% were giving drinks other than water and milk.

All items had percentage agreement over 70% for both reliability and validity. There was however not much variation in the responses hence kappa values were low for some of the items.

Seventy percent of the mothers agreed or strongly agreed with the statement that 'it is possible to feed a baby too little'; whereas only 50% agreed or strongly agreed with 'it is possible to feed a baby too much' (p = 0.002). Ninety percent were confident that they could feed their baby so that they gain enough weight; whereas 72% were confident that they could feed their baby so they do not gain too much weight (p < 0.001). Ninety-eight percent perceived that their baby's weight was about right.

The 11 items in this domain measured three constructs - Outcome-expectancies (5 items), Self-efficacy (4 items) and intentions (2 items). All items had percentage agreement over 76% for reliability and 82% (9/11) items had percentage agreement over 66% for validity. The two items that had percentage agreement below 66% (56% and 64%) had kappa values of 0.09 and 0.22.

As quantified by responses to the first questionnaire, mean self-efficacy score was 2.7 (median 3, inter quartile range 2.6-3.3); for outcome-expectancies mean score was 2.7 (median 3, inter quartile range 2.6-3.4); and for intentions mean score was 2.6 (median 3, inter quartile range 2-3.5), on a scale of 1-5. Internal consistency measured by Cronbach's alphas was 0.51, 0.79 and 0.90 for self-efficacy (4 items), outcome-expectancies (5 items) and intentions (2 items) respectively. Correlations between summary scores for self-efficacy, outcome-expectancy, and intentions were 0.84-0.92 (all p < 0.0001-Table 5),

^aTest-retest reliability-1^st and 2^nd questionnaires

^bValidity-1^st questionnaire and interview

^cFrequency of respondents 1^st questionnaire - for Question 11a- answering scoops rounded/heaped; 11b loosely packed; 11c added powder first. For Question 12a-12 g answering followed all or some of the advice.

^dQuestion- Question number

^eAgreement- Percentage exact agreement- over 66% good agreement

^fKappa- Chance corrected agreement or Cohen's Kappa- < 0 poor; 0-0.20 slight; 0.21-0.40 fair; 0.41-0.60 moderate; 0.61-0.8 substantial; 0.81-1 almost perfect

^aTest-retest reliability-1^st and 2^nd questionnaires

^bValidity-1^st questionnaire and interview

^cQuestion- Question number

^dCorr-Spearman's correlation coefficient

^e%Ag- Percentage exact agreement- over 66% good agreement

^fKap- Chance corrected agreement or Cohen's Kappa- < 0 poor; 0-0.20 slight; 0.21-0.40 fair; 0.41-0.60 moderate; 0.61-0.8 substantial; 0.81-1 almost perfect

^aTest-retest reliability-1^st and 2^nd questionnaires

^b1^st and 2^nd Questionnaires coded as positive, or neutral, or negative

^cValidity-1^st questionnaire and interview coded as positive, or neutral, or negative

^dFrequency of respondents 1^st questionnaire

^aTest-retest reliability-1^st and 2^nd questionnaires

^b1^st and 2^nd Questionnaires coded as positive, or neutral, or negative with negatively worded items reverse coded

^cValidity-1^st questionnaire and interview coded as positive, or neutral, or negative with negatively worded items reverse coded

^dFrequency of respondents 1^st questionnaire with negatively worded items reverse coded