I am running a multinominal logit regression with five dependent variable categories: "I disagree fully", "I disagree", "Neutral", "I agree" and "I agree fully". I computed marginal effects in Stata (margins dy/dx in Stata), which show the difference in probability of each of the dependent variable categories associated with a one unit change in each of the independent variables. Only the marginal effect for "Neutral" is significant and negative for one of the independent variables. If the mean of the dependent variable is "I agree", can I then conclude that the independent variable has a positive effect on the dependent variable?