I’m currently trying to test if differences in proportions of people infected by malaria (RDT positive) between clusters with high or low coverage of control intervention are significant. Therefore I’m running a quasi-binomial GLM in R like this:

```
fit <- glm(cbind(RDT_pos, RDT_neg)~ Coverage_ov75 + rur_urb + pattern, data=df, quasibinomial)
```

My explanatory variable of interest is `Coverage_ov75`

; the others are controlling variables. Everything is ok until I test the significance. When I use `summary(fit)`

, p-value is 0.0630, so the result is non significant at .05 level:

```
Call:
glm(formula = cbind(RDT_pos, RDT_neg) ~ Coverage_ov75 + rur_urb +
pattern, family = quasibinomial, data = df)
Deviance Residuals:
Min 1Q Median 3Q Max
-3.7111 -1.9594 -0.7794 1.0525 3.8943
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -5.1581 0.7147 -7.217 7.48e-07 ***
Coverage_ov751 -1.4263 0.7224 -1.975 0.0630 .
rur_urb1 0.3906 0.5999 0.651 0.5227
patternHighlands 1.3138 0.7581 1.733 0.0993 .
patternSouth 1.6602 0.7000 2.372 0.0284 *
patternWest 1.3084 0.7271 1.799 0.0878 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for quasibinomial family taken to be 5.131962)
Null deviance: 157.92 on 24 degrees of freedom
Residual deviance: 103.30 on 19 degrees of freedom
AIC: NA
Number of Fisher Scoring iterations: 5
```

But when I use odds ratios’ 95% confidence intervals, the result becomes significant –i.e. 95%CI excludes 1.00 value-, whether I use `confint`

or `confint.default`

function:

```
> exp(cbind(coef(fit), confint(fit)))
Waiting for profiling to be done...
2.5 % 97.5 %
(Intercept) 0.005752403 0.00115120 0.01973567
Coverage_ov751 0.240198126 0.04344154 0.83777344
rur_urb1 1.477932315 0.47495969 5.29358843
patternHighlands 3.720230839 0.83163497 18.19418962
patternSouth 5.260253952 1.37407939 23.72815126
patternWest 3.700344766 0.91283718 17.44061487
> exp(cbind(coef(fit), confint.default(fit)))
2.5 % 97.5 %
(Intercept) 0.005752403 0.001417273 0.02334774
Coverage_ov751 0.240198126 0.058304544 0.98954790
rur_urb1 1.477932315 0.456059336 4.78947312
patternHighlands 3.720230839 0.841872843 16.43967686
patternSouth 5.260253952 1.334015707 20.74208834
patternWest 3.700344766 0.889853675 15.38741905
```

The result from confint.default is very close to significance level but the p-value is not so close to .05, at least not as much as what I previously observed in discrepant results (or as others describe e.g. in Differences between conclusions from a p-value and confidence intervals) .

I’m not used to quasi-binomial models, so I wonder if my procedure correct. If it’s correct, than how should I interpret these discrepant results?

Many thanks in advance.